A Microfluidic Device for Automated High Throughput Detection of Ice Nucleation of Snomax®

Abstract

Measurement of ice nucleation (IN) temperature of liquid solutions at sub-ambient temperatures has applications in atmospheric, water quality, food storage, protein crystallography and pharmaceutical sciences. Here we present details on the construction of a temperature-controlled microfluidic platform with multiple individually addressable temperature zones and on-chip temperature sensors for high-throughput IN studies in droplets. We developed, for the first time, automated droplet freezing detection methods in a microfluidic device, using a deep neural network (DNN) and a polarized optical method based on intensity thresholding to classify droplets without manual counting. This platform has potential applications in continuous monitoring of liquid samples consisting of aerosols to quantify their IN behavior, or in checking for contaminants in pure water. A case study of the two detection methods was performed using Snomax^® (Snomax International, Englewood, CO, USA), an ideal ice nucleating particle (INP). Effects of aging and heat treatment of Snomax^® were studied with Fourier transform infrared (FTIR) spectroscopy and a microfluidic platform to correlate secondary structure change of the IN protein in Snomax^® to IN temperature. It was found that aging at room temperature had a mild impact on the ice nucleation ability but heat treatment at 95 °C had a more pronounced effect by reducing the ice nucleation onset temperature by more than 7 °C and flattening the overall frozen fraction curve. Results also demonstrated that our setup can generate droplets at a rate of about 1500/min and requires minimal human intervention for DNN classification.

Keywords: Snomax®; automated detection; deep neural network; high-throughput; ice nucleating particle; machine learning; microfluidic device; polarized light.

Figure 8

(a) Plot showing the frozen fraction of Snomax^® in our microfluidic setup vs. experimental data from literature [42,43,45,46,47,50,60,61,62]. The horizontal error bars in the data represent the combined temperature uncertainty of the droplets due to the temperature variation along and across the isothermal channel during experiments. The vertical error bars represent the standard deviation from three independent samples; (b) Plots showing the ice nucleation site density per unit mass of Snomax^® and comparison with literature data [36,42,43,47,50,60,61].

Figure 1

Temperature-controlled microfluidic platform: (a) Schematic showing the complete platform with multiple copper blocks for individually controllable cold zones and the cooling blocks underneath which act as heat sinks for the cold zones; (b) CAD model showing the platform on a microscope stage with a dry, optically transparent enclosure on top surrounding the region where the microfluidic device is placed.

Figure 2

(a) Custom designed control circuit for the temperature-controlled platform (b) Cold zone temperatures measured using thermocouples and microfluidic channel temperatures measured using on-chip thin film sensors. The number labels next to the orange line indicate the corresponding cold zone in the schematic, and the temperature values are measured using thermocouples inserted into the copper blocks. The blue line indicates the temperature in the flow channel. Also shown is the location of the detection region of interest (ROI) during droplet freezing experiments.

Figure 3

PRTD array fabrication and integration with microfluidic poly-dimethylsiloxane (PDMS) channels. The design is based on Stan et al. [48]. (a) PRTD array fabrication and bonding process schematic; (b) Array mask design, showing 19 PRTD arrays in a row.

Figure 4

(a) Microfluidic channel design for the study; (b) Droplet generation at the flow-focusing junction; (c) Droplet ice nucleation and complete crystallization observed in the cold zone. The white region in the background is the temperature sensor underneath the flow channel.

Figure 5

Experimental setup and freezing detection algorithm using a polarized intensity threshold method: (a) Polarized imaging optical path; (b) Region of interest highlighted in blue inside the flow channel; (c–h) Liquid, frozen (bright) and frozen (dark) droplets as they appear inside the region of interest for different analyzer angles indicated.

Figure 6

Experimental setup and freezing detection algorithm using a deep neural network: (a) Bright-field imaging optical path; (b) Region of interest highlighted in yellow inside the flow channel; (c–f) Liquid and different frozen droplets as they appear in bright field images.

Figure 7

Comparison of the different methods for detecting frozen droplets. The dotted line denotes an ideal detection method i.e., where the method performs identical to a human operator.

Figure 9

(a) FTIR spectra of D₂O, untreated, 55 °C and 95 °C treated Snomax^® samples in the Amide-I region; (b–d) Peak resolve analysis of the amide-I region for distinguishing secondary structure of the ice nucleating protein inaZ in Snomax^® with the heat treatment conditions for the samples indicated in the image.

Figure 10

(a) A plot of the frozen fraction of heat treated Snomax^® samples as a function of freezing temperature; (b) A plot of the frozen fraction of room temperature aged Snomax^® samples as a function of freezing temperature.