Optically Induced Dielectrophoresis and Machine Learning Algorithms for the Identification of the Circulating Tumor Cells ^†

Abstract

Detecting circulating tumor cells (CTCs) is a challenge in cancer research. Their dissemination into the blood stream represents a crucial event in the formation of the metastases from the primary tumor. For this reason, targeting CTCs in human liquid biopsies is a warning event for cancer invasiveness, progression, and prognosis. In this regard, by means of the optically induced dielectrophoresis (ODEP) technique, we investigated the response to the electric field, at different frequencies, of human prostatic carcinoma PC3 cells, which mimic CTCs derived from prostate cancer, and human leukemia monocytic THP-1 cells, which simulate circulating monocytes. The obtained spectra of the cell motion descriptors represent the unique identification signature of each cell type.

1. Introduction

The high-purity isolation and analyses of CTCs are crucial in developing targeted therapy [1]. Additionally, these cells are rare and heterogeneous, so identifying them is difficult. Their detection with common epithelial markers that are not expressed in normal blood cells can fail when some cell subtypes miss the markers mentioned above [1]. In this regard, we explored the potential of using ODEP-based multi-spectral analysis to distinguish between PC3 [2] and THP-1 cells [3] in a label-free manner. The proposed method highlights the dielectric signature of every cell population, allowing for the identification of different cell types by exploiting their response to the DEP force.

2. Materials and Methods

Cell manipulation was performed using an ODEP-based Lab-on-Chip device and an optical platform [4], as shown in Figure 1. An AC potential was applied between the ITO layers, and a light pattern was projected on the a-Si. The generated non-uniform electric field induced a dielectrophoretic (DEP) force on the cells by producing three different cell motions: attractive (pDEP), repulsive (nDEP), or steady-state cell motions. Single cells were measured at different frequencies in a range 25–150 kHz with steps of 5 kHz and an amplitude of 10 Vpp. Videos of the DEP-induced motion were acquired, and the cell tracking procedure allowed us to extract a set of cell parameters for each n frequency: the cumulative displacement $CD\left(f_n\right)$, the maximum velocity $v_{max}\left(f_n\right)$, and the maximum DEP force F_{DEP_max}(f_n) were used as inputs to build a classification model [4].

3. Discussion

The response spectra of CD(f_n) and F_{DEP_max}(f_n) for both PC-3 and THP-1 cells are shown in Figure 2a and Figure 2b, respectively. As shown, the two cell populations exhibited different crossover frequencies: 30 kHz for PC-3 (blue dotted line) and 50 kHz for THP-1 (green dotted line). The extracted features have been used to build an LDA classification model whose confusion matrix and total accuracy value obtained in the 5-fold cross-validation are shown in Figure 2c. The high discrimination rate suggests that the proposed system, with the appropriate operating conditions, can sort cells and isolate CTCs in a label-free manner under continuous flow.