Cancer Holography for Personalized Medicine

Abstract

Digital holography can measure the 3D physiology and motion of cancer cells, allowing identification of effective chemotherapies for patients.

The overarching characteristic of animate matter is that it moves. Living systems move at all scales, including motions that run rampant within cells. And cancer, as a dynamic disease, is characterized fundamentally by altered states of that intracellular motion.

Detecting variations in slow intracellular movements—some as small as nanometers per second—is a technological challenge. In the past ten years, holography has helped to meet this challenge through digital holographic imaging techniques that include quantitative phase imaging (QPI) of isolated cancer cells; 3D tracking of cancer cells migrating through a biological matrix; and biodynamic imaging of biopsied cancer tissues to measure subtle changes in intracellular motions in response to anticancer drugs. These techniques draw on the partial coherence of light to map the 3D properties of cancer cells and tissues, and thereby to better understand the disease and care for patients.

Digital holography of cancer cells

Thin biological specimens are essentially transparent, with little light scattering. That makes it difficult to observe cells in high contrast, either in transmission or reflection, using conventional microscopy. Specimens do, however, significantly modify the phase of transmitted and reflected waves; these scattered fields are the basis of phase-contrast microscopy, which converts phase to amplitude by interference of scattered light fields with transmitted fields inside microscope optics.

Holography—invented by Denis Gabor as a method to correct aberrations in electron microscopy, and later translated to light imaging—similarly uses interference between scattered light and a reference field, and it has great flexibility in the manipulation and properties of the reference field. QPI, for instance, can use holographic interferometry to measure the refractive-index profiles inside isolated cells and to track changes in internal cell morphology and intracellular motion.

The holographic approach to QPI uses an off-axis reference wave to produce an interference fringe pattern on a digital light sensing array, such as a CCD or CMOS camera chip. The technique is called digital holography, as the hologram is recorded digitally and images are reconstructed using computational algorithms. Holograms are famous for providing partial 3D information about a target, but to achieve quantitative 3D performance requires that multiple holograms be acquired, either at successive depths inside a sample or via multiple illumination angles.

Used primarily in transmission, and on isolated cells or small groups of cells, QPI relies on translucent targets to modulate the phase of the transmitted field that interferes with the reference field. By numerically inverting the holograms from illumination at multiple angles, a full 3D quantitative map of the refractive-index variations inside a cancer cell can be reconstructed.

Holography is well-suited for 3D cell tracking of cell movement, such as metastatic cancer cell migration, through translucent 3D biological matrix samples. Biological matrices are composed of fibrous elements like collagen and are translucent, making them available for transmission studies. The 3D reconstruction capabilities of digital holography make it possible to track and measure cell-crawling behavior that can be related back to the biological and genetic characteristics of the cancer cell. For example, the generation of locomotion forces by a crawling cell, coupled to the elasticity of the matrix, creates local deformations that can be measured holographically in 3D, in addition to cell size, mass and morphology.

Migrating melanoma and breast cancer cells have been used to better understand how cancer cells migrate to new locations in the body to form metastatic cancer sites. In this cell-tracking application, digital holography can be performed with a wide field of view to allow many crawling cells to be tracked in 3D at the same time. However, although single cells or 3D biological matrices are transparent and can be imaged in transmission, dense layers of cells thicker than even a tenth of a millimeter no longer transmit much light, and other methods are needed.

Probing cancer tissue dynamics

Thick cancer tissues, such as biopsy specimens, are translucent but attenuate light, preventing their interrogation with transmission holography. Backscattering of light from thick tissues can be relatively bright, allowing the use of light-ranging techniques to measure the optical path distance that the light has traveled.

Although time-of-flight measurements of reflected light pulses (as in conventional lidar) are not practical, optical path-length differences are measured easily using a coherence-gated approach. Coherence-gated digital holography matches the optical path lengths of light scattered from tissue relative to a second, delayed reference wave. Off-axis coherence-gated digital holography measures spatial interference on a pixel array by imaging holographic fringes at a fixed path length. This contrasts with optical coherence tomography (OCT), which uses rapid path-length scanning or swept sources.

A convenient configuration for coherence-gated digital image reconstruction places the pixel array on a Fourier plane of the imaging optics. The light scattered from a biological sample is transformed through a 4f configuration to a Fourier plane, which is then de-magnified through a third lens onto the CCD or CMOS array, where the signal intersects the reference wave in an off-axis condition. The demagnification factor and off-axis angle are set by the “rule of nine” condition for practical digital holography: three camera pixels per fringe by three fringes per speckle. The Fourier-domain hologram is reconstructed to the image domain using efficient fast Fourier postprocessing. The low-coherence light source, typically a superluminescent diode (SLD), creates interference fringes from an optical section that is approximately as thick as the coherence length of the light source, although in practice it is broadened by multiple scattering.

One of the chief differences between low-coherence digital holography and conventional OCT is that the former’s broad-area illumination, without spatial filtering, permits multiply-scattered light to contribute to fringe contrast. This produces well-developed speckle that is spatially modulated by the holographic fringes and temporally modulated by intracellular motions.

Image reconstruction using the fast Fourier transform produces images that have fully developed speckle without spatial features. In OCT applications, this is called channel cross-talk and is considered a parasitic effect. For biodynamic imaging, however, it has an advantage: The speckle—created by compounded multiple phase excursions caused by multiple light scattering—is highly sensitive to minute phase changes induced by intracellular transport. This forms the basis of a highly effective approach to measuring motion in the cell, Doppler fluctuation spectroscopy.

Intracellular Doppler spectroscopy

The speeds of cellular and intracellular components inside living tissue range from tens of nm per second (slow membrane shape changes and cell motility) to tens of μm per second (fast organelle and vesicle traffic inside a cell). In the near IR using a light-backscatter configuration, these speeds produce Doppler frequency shifts from 10 mHz to 10 Hz. At the ultra-low-frequency range of 10 mHz, this represents an impressive frequency shift of one part in 10¹⁶ on the central frequency of the probe light. This extreme sensitivity is beyond the capabilities of direct frequency measurement, but is possible through interferometric phase-sensitive detection. The most sensitive approach to characterize intracellular dynamics is through high-dynamic-range speckle fluctuations.

The active motions inside cells produce dynamic light scattering signals that mimic diffusion, but that have characteristic frequencies ω_d = 16π²n²v₀²τ/λ₀² that are related to the average speed v₀ of the scattering elements and the transport mean-free time τ, where λ₀ is the free-space wavelength and n is the refractive index of the tissue. The biological processes inside cells are actively driven and highly processive—meaning that motions are directed and persistent, though randomly oriented. The persistence lengths of most intracellular transport exceed the reduced wavelength, λ_red = λ₀/4πn, by an order of magnitude or more, placing the light scattering firmly in the Doppler regime. Therefore, fluctuation spectroscopy measures the combination of beat frequencies among all the different Doppler frequency shifts caused by the many transport processes in the cells.

A typical time-domain digital holography data set consists of holograms captured at a frame rate of 25 fps (Nyquist frequency of 12.5 Hz) with a base frequency of 6.3 mHz. Image reconstruction is performed during post processing of the captured holograms, and the intensity fluctuations of single pixels are Fourier transformed and averaged to generate fluctuation spectra for the tissue. Multiple sets can be acquired for longitudinal studies of tissue dynamics using tissue dynamics spectroscopy, which can run for hours or days after the tissue is exposed to therapeutic agents such as drugs that affect the cellular cytoskeleton. The fluctuation spectra span three orders of magnitude in dynamic range, and drug-induced changes are typically in the range of tens of percent.

A drug’s effects are represented by spectrograms—time–frequency maps of the relative changes in spectral density. These tissue-dynamics spectrograms show the logarithm of the spectral power densities relative to a baseline fluctuation spectrum that is established in the first three hours. At time t = 0, a therapeutic drug is applied to the living specimen, inducing a change in the power spectrum; increases in spectral density are represented by warm colors (yellow and orange), decreases by cool colors (teal and blue). The spectrogram is averaged over a full coherence-gated optical section near the midpoint of the biopsy sample nearly half a millimeter deep inside the living tissue sample.

These spectrograms can be broadly interpreted in terms of the biological functions that relate to different frequencies. Spectral features at frequencies near 10 mHz, for example, are tied to cell-shape changes caused by processes associated with cell growth and death. This frequency range is primarily rheological, as cells respond to changes in their force environment. On the other hand, the mid-frequency range, between 100 mHz and 1 Hz, is associated with motions of large internal structures, such as the nucleus, and of the cell membrane. Active cytoskeletal reorganization occurs in this frequency range, and inhibited cellular metabolism is reflected in a decrease in spectral density. The frequencies between 1 Hz and 10 Hz are associated with the transport of small internal structures (organelles and vesicles) that are often enhanced during programmed cell death.

Personalized medicine

Chemotherapy drugs that stop cancer cells from dividing, or that induce cell death, cause subtle changes in the intracellular dynamics of cancer tissues; hence, intracellular Doppler spectroscopy can test how well a patient may respond to their prescribed chemotherapy. The test begins by acquiring a cancer patient biopsy at the time of diagnosis, and keeping a small sample of the tissue alive and healthy for Doppler profiling. The sample is diced into multiple sections, each approximately 1 mm³ in volume; different sections receive different anticancer treatments. The fluctuation spectra then measure the tissue response to the therapeutics. The use of intact 3D tissue biopsies, rather than isolated patient-derived cells, is necessary to maintain the natural tumor microenvironment, which helps to regulate how cancer cells respond to applied drugs.

This process can form the foundation for a variety of powerful approaches to visualizing and categorizing patients’ potential drug responses—and tailoring treatment accordingly (see “From Doppler spectroscopy to personalized medicine,” p. 48). Characteristic spectrogram signatures, for example, can flag individual patients as potentially resistant to a prospective therapy, allowing alternative treatment options to be explored. Machine learning can be leveraged to classify patients into potentially drug-sensitive versus drug-resistant cohorts, and to visualize networks of patients with common drug-response characteristics. All of this allows patients to be steered away from ineffective therapies and toward potentially more effective approaches, saving time and money and improving chances for survival—an indication of digital holography’s potential to impact cancer care.

From Doppler spectroscopy to personalized medicine.

The tissue dynamics spectrograms from intracellular Doppler spectroscopy form the foundation of a powerful toolkit for predicting patient outcomes from different cancer chemotherapies.

1. Drug-Response Spectrograms

Drug-response spectrograms (from a trial of an ovarian-cancer drug) show patterns of different drug response for patients who relapsed after treatment (resistant phenotype) versus patients who achieved complete remission (sensitive phenotype). The bottom row shows the difference of the two response groups. The characteristic spectrogram signatures can flag individual patients as resistant to a prospective therapy.

2. Features

The drug-response spectrograms can be used to train machine-learning classifiers to predict patient outcomes. These can take the form of feature vectors (such as those shown here from a breast cancer drug trial in 43 patients) showing shifts or enhancements in different frequency components related to parameters such as cell death, organelle motion or average intracellular speeds. Here, one row is shown for each patient, grouped in resistant and sensitive cohorts.

3. Patient Similarity Matrix

Feature vectors allow calculation of a similarity matrix, based on the “distance” of each patient’s feature vector from one another. The distance measures can be Euclidean distance, correlation coefficients (shown here) or other kernel-defined distances. The patients are grouped into resistant or metastatic (upper) and sensitive (lower) types, and the matrix’s nearly block-diagonal character shows strong intragroup similarity and strong intergroup dissimilarity.

4. k-Neighborhood Network

From the similarity matrix, in turn, a k-neighborhood network is built by creating links to the (k = 4) most similar patients for a given patient. The network graphically separates into relapse and remission groups, with only a few outlier patients who fail to classify correctly. These outliers may have rare cell behaviors that lie outside the binary relapse–remission classifier. When a patient is predicted to fall within the group with incomplete response to a prescribed treatment, alternative therapies can be considered.

Based on these ongoing research programs, and on related advances in the field of holographic imaging, the future of digital holography for cancer research is likely to focus increasingly on dynamical properties of cancer cells and tissues. Motion is such a ubiquitous aspect of living matter, and it is so specific to biological function, that sensitive measurements of motions, and changes in those motions caused by external interventions, is likely to become a mainstay of biological optics.

A 3D reconstruction of a small living cancer tumor, acquired using low-coherence digital holography that performs laser ranging up to 1 mm deep into tissue, is color-coded for intracellular Doppler activity, displaying the high activity (red) of dividing cells and low activity (blue) in the low-oxygen core. Adapted from Z. Li et al., Appl. Opt. 60, A222 (2021)

Left: In digital holography, a pixel array (CCD) records the interference fringes of light scattered from a biological sample interfering with a reference wave (off-axis angle exaggerated) to form a speckle hologram. A phase image of the sample is numerically reconstructed from the hologram. Right: Quantitative phase imaging uses a galvo mirror to change the illumination angle on the sample to generate a series of holograms, allowing the numerical reconstruction of the 3D refractive-index variations in the sample (shown on the right for a live HeLa cell).

Optical schematic of the biodynamic imaging system in a Mach-Zehnder configuration. Light, backscattered from living tissue, is Fourier-transformed onto a pixel array, where it intersects a reference beam in an off-axis digital-holography configuration.

A digital hologram with holographic fringes modulating spatial speckle (left); its fast Fourier transform, showing the side-band images of the tissue section (center); and non-zero-path (NZP) subtraction, which removes the zero order.

Optical coherence image (OCI) of the midsection of a 600-μm-diameter tumor spheroid; a motility contrast image (MCI) representing the time-dependent speckle contrast; and volumetric MCI acquired by successive depth gates.

Time-series analysis of speckle.

1. Time series

At each pixel of the reconstructed image, intensity fluctuations occur across the imaging time.

2. Power spectra

The intensity time series are Fourier transformed into power spectra characterized by a knee frequency (characteristic speed), a slope parameter (persistence length) and a Nyquist floor (fast organelle transport).

3. Tissue dynamics spectrogram (TDS)

In tests of a drug’s effectiveness, multiple data sets, providing information on changes in spectral density across hours or days after the drug is applied, can be combined into tissue dynamics spectrograms plotting the logarithm of the Doppler frequency against the time duration of the experiment. (Shown here is a spectrogram for the cancer chemotherapy drug doxorubicin, which inhibits how DNA is copied in a cell prior to cell division.) The spectrograms can be interpreted in terms of the drug’s effect on biological functions at different scales in the cell.