Cellular biosensing using optical spectroscopy
Techniques such as optical nanoscopy1 and spectroscopy2 have expanded our capability to probe biological cells at the nanoscale. These techniques provide the opportunity to learn about disease mechanisms at the cellular level, and therefore to improve disease detection and diagnosis. Now we have developed approaches for cellular biosensing that use optical information at the nanoscale combined with spectroscopy to discriminate disease.
Phase imaging has long been a mainstay of cell biology, with most laboratories using phase contrast imaging for routine cell visualization. However, phase contrast is a qualitative technique, making quantitative comparison difficult across individual cells. Using interferometry techniques, such as digital holography, it is possible to measure the complex optical field that interacts with a cell to recover quantitative information about its features at the nanoscale. Indeed, many efforts have applied such quantitative phase imaging (QPI) techniques to study cells.3 However, to push the approach toward diagnosis of disease, practical advances are needed.
One way to improve QPI is to incorporate new sources of information. Our QPI approach includes spectroscopy by using a custom spectral filter on the input light (see Figure 1). Light from a supercontinuum source (Fianium SC450) is dispersed using a diffraction grating and then refocused onto the pivot point of a galvanometer scanning mirror, such that different wavelengths are coupled into a single-mode fiber depending on the position of the mirror. Light is available across the range of 475–700nm, selectable in 5nm increments and with a bandwidth of ∼1nm.4 This arrangement offers the benefit of low-coherence imaging, which reduces speckle.
We have applied this approach to several cellular biosensing applications, including imaging of red blood cells (RBCs) and cancer cells. For example, in earlier work, we compared subtle nanoscale changes in cell structure over time of RBCs from patients with sickle cell disease to normal controls, finding differences between the two that indicate changes in mechanical stiffness.5 Although the measurement showed differences, there is a higher bar needed to use the approach for diagnosis: low error rates for both false positives and false negatives are needed. Even with error rates as low as a few percent, a substantial number of patients could receive false results if the test were widely applied.
To advance QPI toward use as a diagnostic method, our more recent work has examined the changes in RBCs due to infection by malaria parasites.6,7 Figure 2 shows typical phase images of uninfected and infected RBCs. Using the spectroscopic capabilities of our imaging approach, we identified changes in RBC properties that correlated with infection stage.6 Analysis of the spectra showed that the hemoglobin content of individual cells was reduced on average by up to 72.2% due to infection. The spectroscopic capabilities also enabled us to make a distinction between hemoglobin content and other protein structures of the cell, using a metric known as optical volume (OV).8 The cells' OV also showed a decrease with disease progression. However, OV was only partially correlated with hemoglobin content, indicating that there is more to detecting this disease than simply measuring the amount of hemoglobin. Overall, the spectroscopic approach produced a low false positive rate of 8.8% but a higher false negative rate of over 20%, indicating that better methods are still necessary.
To improve our ability to diagnose malaria infection, we used machine learning algorithms to automatically analyze phase images.7 In this approach, each phase image of an individual cell is characterized by 23 morphological metrics, including OV but also parameters that describe the symmetry and density variations. By compacting high-resolution data from individual cells into these metrics, we were able to study hundreds of cells within each group and train computer algorithms to identify those infected with malaria parasites. The algorithms produced excellent discrimination, resulting in error rates of < 1%. In some cases, zero false positives occurred and the false negative rate was as low as 3–6 cells per 1000. While further validation is needed to bring this approach into use as a clinical diagnostic tool, these initial results are an important first step toward realizing the accuracy needed for this task.
Finally, we have also applied QPI to detect changes in the mechanical properties of cells that might indicate that they have the potential to become invasive cancer. By flowing fluid past cells, we can induce subtle mechanical changes that we can measure with QPI (see Figure 3). Using various models of cancer cells, we found that we could detect nanoscale deformations that distinguished cells based on their mechanical properties.9 Although this work is at an early stage, the sensitivity of the approach suggests that it has significant potential for use as a diagnostic.
In summary, we have applied QPI to obtain nanoscale information about individual cells and used it for disease diagnosis. The ability to characterize cells at the individual level provides new, more powerful ways to discriminate cells based on morphology and content. Our future research will continue to refine these techniques for high-throughput applications based on robust instrumentation for potential application at the point of care.
Adam Wax is the Theodore Kennedy Professor and is director of graduate studies in the Department of Biomedical Engineering. His research interests are in the use of light scattering and interferometry for diagnosis of disease and fundamental cell biology studies.
