Ovarian Cancer’s Early Warning System

Ovarian cancer has the highest mortality of the gynecological cancers; in the UK alone, there were more than 7,000 diagnoses in 2010, and over 4,000 deaths in 2011. These statistics are partly a consequence of the disease being largely asymptomatic in its early stages; most patients are diagnosed late, when the five-year survival rate is less than 40 percent. If we compare this with the five-year survival rate for early diagnoses – 90 percent – the need to improve screening and early detection becomes obvious. We’ve made it a mission to identify biomarkers to hopefully address this dismal outcome.

A few biomarkers for ovarian cancer already exist, but none suitable for screening. This is partly because of methodological issues: many studies have relied on patients who had already been diagnosed. This approach is good for identifying markers concurrent with clinical disease, but not so good for identifying those present at the early, asymptomatic stage. Clearly, to find precise early-stage markers requires resources, rigorously designed studies, and the ability to process and store the reams of data that you’re going to amass (see Box “Criteria for Studies to Identify Biomarkers”). And in those respects, I’ve been very fortunate.

A question of resource

I work with the UK Collaborative Trial of Ovarian Cancer Screening (UKCTOCS), which received £26 million in funding from the Medical Research Council (MRC), Cancer Research UK, and the Department of Health, with additional support from the Eve Appeal, Special Trustees of Bart’s and the London, and Special Trustees of UCLH. These invaluable funds have allowed the enrolment of 202,000 women in a study spread over 13 centres. And we’ve followed them for more than 10 years. What’s more, 55,000 of these women gave serum samples each year for the 10-year period, so we have access to a very powerful sample set which allows us to track proteins of interest right back to a timepoint of several years before diagnosis. To have serial serum samples from 55,000 individuals over that kind of time period is a tremendous resource! And we also have support from the PROMISE program (Predicting Risk of Ovarian Malignancies, Improved Screening and Early detection), funded by CRUK and the Eve appeal, which is aimed at halving the number of deaths from ovarian cancer. We’re very pleased to be so well-resourced.

That said, working with serum is extremely difficult! The most abundant 22 serum proteins account for about 99.9 percent of the protein mass in serum – so they block out everything else. This is a problem, because traditional mass spectrometry uses Data Dependent Acquisitions (DDA), which focus on only the most abundant ions; these are selected for fragmentation, leaving many less abundant ions unexplored. Consequently, all the information pertaining to the lower-abundance serum components is lost. For that reason, we’re employing immunodepletion of the abundant proteins and Data Independent Acquisitions (DIA), in which all of the potential ions are fragmented. DIA is one of the key technologies we have where I work at the Stoller Biomarker Discovery Centre; in particular, we’re using Sequential Window Acquisition of all Theoretical fragment-ion spectra mass spectrometry (SWATH).

Taking the independent approach

How does SWATH do it? It rapidly cycles through sequential isolation m/z windows over the whole liquid chromatography elution range. Importantly, it generates a complete – and permanent – record of the fragment ion spectra of all the components in a biological sample, within a predetermined m/z versus retention time window. This dataset can be re-mined time and again based on new hypotheses. Simply put, SWATH provides us with a permanent digital record of every single biological element in that sample. This is really important, because the problem with serum biobanks is that once the sample’s been used, it’s gone forever! By contrast, after we’ve translated the physical biobank into a digital dataset, we can return to it time and again to re-mine it for new information. We can transport it with us where we please, or send it to other labs so that they can work with it too.

SWATH is a very powerful technology, but how is it used in practice? Our method is to synthesize peptides that are representative of proteins suspected to be early markers of ovarian cancer. Then we run them through a mass spectrometer and use the result to interrogate the SWATH maps. Basically, we develop a spectral library comprising information on the different peptide masses, retention times and intensities, and this allows us to quantitate protein changes in the serum samples. These data inform the next stage of the biomarker validation process.

The importance of Protein Z

We used that exact method to investigate markers for the two forms of ovarian cancer: Type I, which is comparatively slowly progressing, and Type II, which is more aggressive. We analyzed 500 serial serum samples from Type I and Type II patients, spanning a period of seven years pre-diagnosis, and compared levels of Protein Z with those of CA125, which is the gold standard biomarker for ovarian cancer. When we imposed spectral libraries on SWATH maps that we’d generated from serum samples, we found that Protein Z is significantly downregulated in Type I ovarian cancer patients. Furthermore, reduced serum levels of Protein Z are evident not only near to diagnosis but also – and more importantly – a long time before diagnosis. We confirmed this pattern with non-linear modelling. Interestingly, we found that Protein Z levels in Type I patients decrease at the same point that CA125 levels go up. In fact, our data suggested that using Protein Z serum levels as a diagnostic tool captures a number of patients who would have been missed using CA125 alone.

When we looked at Type II patients, we again found that Protein Z serum levels change significantly with disease progression; however, the change is in the opposite direction – serum levels increase. Again, using Protein Z levels as a triage mechanism captures patients that are missed if we rely on CA125 levels.

Protein Z is clearly of interest, but is a biomarker any use if it goes up in some patients and down in others? Emphatically, yes: measuring both Protein Z and CA125 in Type I and II patients shows that Type I patient levels hardly ever breach the Type II threshold, and Type II patients generally don’t breach the Type I threshold – they are almost mutually exclusive (see Figure 1). So, we believe that Protein Z has real potential in ovarian cancer diagnosis.

Of use for screening?

We’ve now incorporated Protein Z into a panel of markers – which we also identified by SWATH – and built prediction models for Type I and Type II ovarian cancer. For both types, our models give break points significantly earlier (i.e., in advance of diagnosis) than CA125 alone (see Figure 2).

We believe this panel could be the basis of a screen that will pick up ovarian cancer a year or two earlier than the current standard of care. We envisage it being used as a clinical tool in serum sample analysis: measurement of serum levels of these four proteins, followed by application of a simple algorithm, will separate patients into low-risk or high-risk categories and guide them to corresponding management pathways. In the small sample we’ve looked at, this approach seems a lot better than using CA125 alone.

Finally, from a pure technology perspective, I’d expect SWATH to be increasingly used as a diagnostic tool in itself. SWATH maps contain millions and millions of data points, and by application of the right statistical tools we should be able to build algorithms – almost like facial recognition algorithms – and start clustering people into various groups according to their likelihood of developing a given disease. That’s the aim of our research and our realistic hope for the future.

Robert Graham is Senior Lecturer in Clinical Proteomics and Deputy Director of the Stoller Biomarker Discovery Centre, University of Manchester, UK.