Controls are a crucial part of any assay. Without them, our work is, at best, simply a waste of time and reagents. We know we need them but we probably should give them much more thought than we do. CLI caught up with Dr John Hedges, Director of Research and Development at Asuragen (a Bio-Techne brand), to discover more about different types of controls and the characteristics of a good control.
‘Molecular Diagnostics’ is a commonly used term and may mean different things to different people. In order to set the scene, can you give us an overview of what is meant by this term?
Molecular diagnostics involves the detection and/or quantifi-cation of any DNA, RNA, or protein markers that have clinical validity. Example uses include diagnosing disease, monitoring progress, determining prognostic risk, or deciding which therapies would work best for individual patients. For example, the t(9;22) chromosomal translocation results in formation of the BCR-ABL1 fusion gene. Expression of the e13a2 or e14a2 forms of the BCR-ABL1 fusion transcript accounts for approximately 98% of cases of chronic myeloid leukemia (CML). In patients with this type of cancer, moni-
toring the level of aberrant BCR-ABL1 fusion transcript in the blood over time can provide important information about tumour burden, which is associated with effectiveness of treatment and relapse. Molecular diagnostic tests that target this mutation and report levels using the International Scale (which was established through a collaboration between a drug manufacturer and labs specializing in molecular testing for clinical disease management) can deliver highly quantitative readouts, making them essential and regular components in CML patient care.
From an early stage in science education we learn about the need for controls – what are they and why are they needed?
Controls tell us about the performance of the assay workflow. They can also provide truth against which individual patient results can be assigned, such as traceability to a standard reference value. One thing we don’t always learn early in our science education is exactly which kinds of controls are needed and when.
There are multiple categories of controls used in a molecular diagnostics workflow: positive/negative, exogenous/endogenous, and internal/external. Positive controls contain or mimic the analyte of interest and should be detected by the assay if it is set up correctly. If a positive control is not detected or appropriately quantified, then test results for the entire batch are considered invalid. The inverse is true for negative controls, which are designed to be undetectable or a true negative for disease. Both positive and negative controls can be used at multiple points of a molecular assay workflow to control each step of the process, but they are most commonly used at the analytical step that will generate
a patient’s results.
Process controls ride along next to or within the sample from a particular step to the very end. They help to ensure that this portion of the process – not just the analytical portion – operates as expected. For example, an endogenous internal control (natively present in the sample) or an exogenous internal control (added to a sample prior to extraction) can be used to verify that both the pre-analytical and analytical stages of testing were valid. By comparison, external controls that are separate from the samples, but processed in parallel, can verify the entire workflow and offer confidence in all samples in the run.
To reduce the risk of invalidating a true test result, most molecular assays use a combination of controls to enable a matrix of determinations of whether a run or individual unknown result is valid. Let’s take, for example, an assay having positive and negative run controls and an exogenous internal control added to each sample before extraction. If the exogenous positive and negative run controls are valid, but the exogenous internal control is negative, then there is little to no confidence that the biological analyte of interest in the collected sample would have also made it through the extraction process. Therefore, a negative result for this sample should be considered inconclusive. However, it is still possible to confidently interpret results for samples in which the exogenous internal control is positive.
Having clinical samples and well-designed controls available for use as early as possible in the assay development process is invaluable in ensuring an assay design meets requirements for accuracy and precision. As part of a risk-based approach for assay development, the most essential risk points can be identified and appropriately controlled and monitored as needed. This can significantly reduce the burden of introducing a control at every single point in the testing process. As an example, if interfering substances can be de-risked as a significant source of variation during development, the inclusion of an exogenous internal control for each sample may be avoided.
What are the characteristics needed for good controls?
An ideal external control would behave identically to the assay’s target sample and analyte type. However, this is not usually practicable. Wherever possible, good controls should behave similarly to clinical samples. This means they should contain the exact sequence and design of the native analyte and contribute samplelike matrix effects to assess the full assay workflow in every run.
Good controls should also be commutable, meaning they behave consistently and are traceable to reference materials when tested in any variety of applications. It is important to maintain this traceability no matter how far removed the control may be from the original reference material source (which is often highly characterized cell lines in very limited supply). For example, the inclusion of calibrators to the International Scale in our previous example of BCR-ABL1 monitoring is critical for ensuring commutability of patient results across testing sites. Valid calibration results facilitate review of control reaction results. In turn, valid control results allow review of measurements from patient samples. Both CLSI EP32 and ISO 17511 [1,2] provide a guide for how this traceability should be established and maintained for in vitro diagnostic medical devices.
Although human reference materials and human cell lines are preferred as external controls given their greatest similarity to clinical samples, they can be very difficult to source and maintain at supply levels necessary for broad distribution. Therefore, commutable synthetic controls are an important part of every laboratory workflow because they offer a reliable, renewable control source that can be tightly controlled at all manufacturing and quality control steps to ensure lot-to-lot continuity and reliability.
Synthetic controls can also be safer to work with in a clinical lab. For certain types of pathogens, using clinical samples as controls means that there is risk of infection to technologists. Synthetic controls are designed to mimic the native analyte as closely as possible to provide a reliable control alternative, but they do not contain the actual pathogen.
Depending on the application, analyte stability may be another important factor to consider. Synthetic controls can be produced in so-called ‘armored’ versions that encapsidate the analyte of interest in a protective shell. This makes the controls more stable for use as process controls, so they can be added at the point of sample collection and extraction as a marker that can be traced along the entire path of the sample to ensure the whole workflow was performed correctly. Liquid samples, such as blood, urine, swab transport media, etc., can be collected directly into tubes pre-loaded with armored exogenous internal controls which can then be co-detected with the analyte of interest. This type of control that rides along with the sample at each processing step is especially important for RNA analytes given the ubiquitous nature of RNases. For quantitative assays, such controls can also be used to normalize results based on the amount of sample potentially lost during the pre-analytical processing steps.
Armored controls are especially good for mimicking viral pathogens; however, such protective coating may not be needed for applications like genetic or oncology testing where lab teams do not need a virallike particle for their workflow – in these cases, well-characterized, unprotected synthetic RNA or DNA molecules often suffice. Across all workflows, synthetic controls used as exogenous internal normalizers should be carefully selected to ensure they do not contain sequence that is homologous to the target human or pathogen DNA.
Finally, appropriate control materials should be the same molecular category as the analyte of interest. For example, an endogenous internal positive control for an assay targeting an RNA from human tissue should also be an RNA. Likewise, external controls for such an assay should be composed of RNA, as use of DNA would not control for the relative batch run inefficiency of the reverse transcription step. This is especially important for quantitative applications such as BCR-ABL1 fusion transcript monitoring where RNA-based calibrators and controls more closely mimic the RNA analyte targeted in the sample in terms of both reverse transcription and qPCR efficiency. On the other hand, DNA-based calibrators and controls only reflect the efficiency of the qPCR step which could reduce the accuracy of sample quantification.
Historically, much research on molecular diagnostics has been done using biobanks and studies primarily involving people of European descent, which we now know can be inadequate for other populations. Does this knowledge also affect the controls we need to use?
As we have collected more genomic information from a broader variety of sources, it has become clear that there are different variants and variant groups (e.g. inversions, deletions, duplications, etc.) that are more prevalent in certain ancestral populations. Indeed, the types of variation can differ widely across ancestral backgrounds. Historically, most of the controls developed have been focused on designs that took their cues from databases that were highly enriched for European ancestry. Because of that, controls have been disproportionately designed for the types of disease variants most commonly found in populations with European ancestry.
As we’re empowered with more data across other ancestral groups, we can design assays for variants that are more important in those populations, and the same is true for controls. There is more information available every single day, so this area is constantly evolving. A major advantage we see with synthetic controls is that they can be more quickly designed and characterized to support variants that are important in a variety of ancestral groups, whether they’re single nucleotide variants, copy number variants, repeat expansions, large insertions or deletions, or other complex structural variants. This is much more challenging to achieve with clinical sample controls or cell line controls, where ensuring a consistent supply is notoriously difficult.
For an example of genetic variation differing by ancestry, consider carrier status for spinal muscular atrophy. The silent carrier genotype (also referred to as a 2 + 0 genotype) – having two copies of the SMN1 gene in cis on one chromosome and zero copies on the other chromosome – is uncommon among people with European ancestry, but much more common among people of African ancestry. To improve detection of 2 + 0 carriers in this higher-risk population, assays have been developed to detect single nucleotide variant markers in SMN1 that co-migrate with the 2 + 0 genotype. Therefore, having the flexibility to quickly generate appropriate controls of such variants for use in quality control testing or as run controls provides a surrogate for limited silent carrier sample availability and helps ensure the reliability and accuracy of testing for multiple ancestries.
To what extent does bioinformatics play a role in assay interpretation now (and envisaged for the future) and how can controls be designed to assist this?
With the evolution of technology – especially massively parallel sequencing techniques such as next-generation sequencing (NGS) – clinical labs are adopting workflows that are more complex and span several days. As this happens, the need for reliable controls is greater than ever. The volume of data generated by NGS-based assays is tremendous, requiring bioinformatic pipelines capable of performing massively parallel analysis. When there are so many steps in a workflow and so much data being produced, it can be incredibly challenging to spot a problem and trace it back to its source. Incorporating controls at every step – including sample preparation, target amplification, library preparation and purification, and sequencing – can aid in detecting where a workflow fails and save valuable time in the troubleshooting process. In addition to physical control materials, in silico controls are also becoming increasingly important in modern NGS workflows to verify and monitor analytical performance when it is not possible to source reference materials for the myriad of variants that are now covered. Bioinformatic pipelines should be built to
integrate control data at every step to streamline troubleshooting and analysis.