Biomarker imprecision in precision medicine

KEYWORDS:

• Biomarker
• oncology
• precision medicine
• theranostics
• genomics
• diagnostics
• personalized medicine

Precision medicine is a conceptual model for disease taxonomy, formally devised in 2011, that extends the long-established clinical pathologic correlation disease model to include and emphasize genetics, genomics, and systems biology knowledge [1]. Precision medicine has the goal to provide more effective therapy for individuals guided by more precise diagnostics. This includes imaging, molecular diagnostics, and other ‘omics’ data as well as the analysis of large data sets by advanced informatics tools including data mining, machine learning, and other aspects of artificial intelligence. Accordingly, precision medicine is highly dependent on the adequacy of biomarkers used to classify disease, determine prognosis, guide treatment, and assess therapy response.

Biomarkers for precision medicine are fraught with both conceptual and implementation challenges. Conceptually, precision medicine is predicated on ideas that application of the vast amount of genomics and other biochemical data generated by life sciences research over the past 60 years, with appropriate detection and data mining, can yield more precise diagnostics and therapies. So far, this hope has been only partially realized and critical reaction is now appearing [2–4]. Despite some important and well-documented successes particularly in cancer [5,6], increasingly, molecular pathways are usually discovered to be more complex upon further investigation. Patients exhibiting similar molecular patterns of disease continue to have a range of responses beyond what the molecular mechanisms would have predicted, and targeted drugs are often effective on only a portion of patients deemed susceptible and often only for short periods, rendering less precision to disease categories at high cost. Genomics has long been known to affect strongly drug responses in certain individuals and families but clinical applications of pharmacogenomics, a clear subject for early deployment of precision medicine, have been slow to develop because of practical implementation difficulties [7,8]. In part, the difficulties with the precision medicine model relates to its overemphasis on genomic changes as disease determinants and lack of adequate recognition that gene expression may or may not translate to protein expression. As well, physiologic reactions and adaptions to molecular changes that counteract specific gene and protein expression can affect disease phenotype profoundly. These difficulties, which impact adversely on biomarker clinical validity, are likely to be resolved only in part over time by advances in understanding basic biology. One promising approach, the Immunoscore, standardizes the assessment of immunologic reaction within the tumor as a prognostic indicator for colon cancer [9]. This suggests that the precision of precision medicine can be improved by assessing key indicators of both the disease itself and the reaction to disease [10,11].

Meanwhile, practical obstacles to biomarker implementation that relate to biomarker diagnostic reliability are the major factors that limit precision medicine advances at present. Many of these factors are addressable by care in applying laboratory diagnostic technologies. These factors include specimen acquisition, biomarker preservation, biomarker analysis, and biomarker informatics.

Specimens for cell, genomic, proteomic, and metabolomic biomarkers are acquired either directly from affected tissue by biopsy usually needle biopsy, or indirectly from blood, ‘liquid biopsy.’ Both methods have limitations of specimen adequacy, and representativeness that limit diagnostic precision. While tissue biopsy provides localization, and with imaging, tissue selectivity, accessing deep tissue sites can be difficult, expensive, and present some patient risk. Further, it is likely that some biomarkers can be degraded during specimen acquisition by insertion force energy such as that induced by core biopsy guns [12]. Alternatively, while intrinsically less invasive, liquid biopsy can be imprecise if marker concentrations in blood are inadequate or not representative of the local lesion.

Variable biomarker preservation is a major source of diagnostic imprecision which is often not sufficiently recognized. Specimen fixation for biomarkers is time dependent, tissue mass dependent as it takes time for fixative to penetrate issue, and analyte dependent. Each biomarker class, DNA, RNA, lipid, protein has different optimum fixation conditions. Particularly for retrospective studies, it is common to compromise fixation by using ultrasensitive probes that can detect and measure biomarkers in the common fixation process for histopathology, formalin fixed, paraffin embedding. Formalin aqueous solution degrades biomarkers as does protein crosslinking, subsequent dehydration, heat, and lipid removal. In this process, genomic markers can not only be lost but artefactual mutations generated [13].

Biomarker imprecision related to lack of analytic rigor is well known. However, the extensive requirement for controls is often not fully appreciated. The need for adequate controls is increased with gene amplification and migration from selected target analytes to genome scale strategies such as DNA or RNA sequencing. Further, each analyte has a residual analytic imprecision; with signature type biomarkers, the imprecision of the biomarker rises proportionately to the number of analytic components within the biomarker [14].

Molecular Biomarker quantitation from raw data often involves application of informatics techniques including advanced statistics and artificial intelligence. Biomarker imprecision is introduced in this step by iterative techniques with limited data sets and controls. In extreme cases, the imprecision leads to data overfitting, calling into question the clinical validity of results.

A hallmark of precision medicine is that genomic biomarker interpretation is often contingent on other information which may be accessible only through networking of multiple specialized data bases [15,16]. While the necessity for these information networks was recognized more than 15 years ago [15], implementation is still a work in progress [16]. This raises questions about data quality and validity that are intrinsic to the biomarker and its clinical utility, questions which increase with the amount and variety of information used as well as the number of databases polled [17].

Several avenues are available for improving biomarker precision. First, the biomarker should be obtained from the most representative site if possible. Only when this is too difficult or when there are positive reasons such as need for broad sampling from many metastatic tumor sites should sampling via liquid biopsies be the method of choice.

To obtain biomarkers from a representative site, imaging guidance, typically ultrasound, or magnetic resonance imaging are employed. Improvements in image resolution contrast, developments in functional imaging, and biopsy needle visibility are all modalities which can enhance lesion localization. To improve needle biopsy performance with existing techniques, competence-based training and quality assurance programs to evaluate and reinforce operator learning are now widely deployed [18]. While mechanical improvements have been made for needle biopsy technology particularly ultrasound-guided endoscopic biopsy, the opportunity remains to devise more ergonomic, more reliable, and less invasive ways of acquiring biopsy samples through needles. These improvements would lower the barriers for needle biopsies facilitating wider and more frequent sampling, thereby raising diagnostic precision.

Timely and appropriate sample fixation is a major domain for biomarker precision improvement [19]. Rather than place a sample in the fixative of convenience at time of convenience, samples should be placed immediately after biopsy in the fixative most appropriate for the biomarker to be assayed. For labile analytes such as RNA, fine needle aspiration biopsy is to be preferred as cells are retrieved intact and as fixative can penetrate individual cells faster than tissue extracted by core or open biopsies. It is also desirable to estimate the adequacy of the biopsy mass during the procedure to ensure that sufficient material is available for reliable analysis.

Regarding analytical precision, robust controls for each analytical step are standard practice and well understood. However, especially for signature based biomarkers, precision can be improved if fewer analytes are used and if informatics technology deployed is transparent for reproducibility.

While current emphasis in precision medicine diagnostics has been centered on cellular analysis of tumors, the concepts apply equally to blood-based markers of disorders such as dementia [20] or spondyloarthropathy [21] where tissue samples are not easily accessible. As well, blood markers may enhance precision medicine if it is desirable to assess the analyte in blood serially. Serum intestinal alkaline phosphatase in inflammatory bowel disease [22,23] and Nerve Growth Factor in treatment of depression [24] are emerging examples.

Biomarker utilization in precision medicine is highly dependent on networked large databases coupled to advanced information science. This has been made feasible by hardware and software advances in computer and communications technology and is driven by commercial and academic interests that are searching for applications for these technologies. Accompanying this trend toward ‘big data’ and ‘deep learning’ are challenges for data validity, data precision, and cost effectiveness [25]. With immense scientific and economic pressures to deploy the feasible ‘big data’ technology, ideas to achieve the reasonable, i.e. cost effective, precise biomarkers based on small amounts of high quality data, are being submerged.

Recognition that widespread application of precision therapeutics is a future aspiration rather than present reality and that biomarker imprecision is a major confounding variable has two important implications for development of precision medicine theranostics. First, increased attention to improve biomarker precision by reducing procedural steps and by optimizing specimen acquisition, preanalytical, analytical, and data analysis for both existing and novel biomarkers, is worthwhile. Second, strategies to optimize present imperfect therapies with existing biomarkers through such tools as mathematical modeling to determine the best modalities for treatment is likely to result in better, more cost effective patient care [26]. In effect, shifting some resources to practical activities that enhance quality of biomarker analysis have the capability of greatly increasing precision in the present state of precision medicine.

Declaration of interest

The author has founder and financial interests in Rna Diagnostics Inc, ProteocyteAI Inc,, Swan Cytologics Inc, and Keyintel Medical Inc (all in Toronto Ontario, Canada). The author has no other relevant affiliations or financial involvement with any organization or entity with a financial interest in or financial conflict with the subject matter or materials discussed in the manuscript apart from those disclosed.

Reviewer Disclosures

Peer reviewers on this manuscript have no relevant financial or other relationships to disclose.

Acknowledgment

The author thanks Heikki Nieminen, John Soloninka, Amadeo Parissenti, and Laura Pritzker, for their helpful discussions.

Additional information

Funding

This paper was not funded.