**3. Results**

#### *3.1. Features of the Hybrid Integrated Biosensor*

The SU8 waveguide (refractive Index ~1.6) integrated well onto the PDMS (refractive index ~1.4) upon curing with UV (Figure 2C). Minor delamination of the SU8 was observed along the walls of the PDMS in some sections, however SU8 was able to guide the input light e fficiently under the PDMS clad. The optical fiber assembly setup (Figure 2D) described earlier provides robust fiber attachment onto the waveguides to enable optical transmission. Also, with this arrangemen<sup>t</sup> of the optical microfluidic chip, the microfluidics and the micromixer modules do not interfere with the functionalities of the optical ensembles, and thus all the components of the system remain independent. Integration of on-chip spectrometer [25] was also tested to conduct direct fluorescence measurement that could be used for specific immunoassays. The fully fabricated hybrid integrated optical microfluidic device is as shown in Figure 2E.

Initial vibration analysis experiments for the mechanical characterization the piezo-actuation system was conducted using laser doppler vibrometry. The natural frequency of oscillation of the piezo-actuator was ~70 kHz, much higher than the intended operating frequencies of the micromixer. As expected, the acoustic waveform generated by the piezo-actuation depends on the operating frequency. For this device configuration, acoustic cavitation inception occurred for operating frequencies greater than ~800 Hz. Upon cavitation inception, the rate of cavitation bubble growth and collapse was dependent on the operating frequency, and these parameters stabilized at ~3000 Hz, beyond which no significant di fferences were observed in the life-cycle or the behavior of the cavitation bubbles. On the other hand, at lower operating frequencies (1~100 Hz), the piezo-actuation principle could be used for transporting fluid across the channels through valveless micropumping [24]. Slight modifications to the existing design enable the feasibility of integrating other types of optical detection techniques, namely fluorescence or evanescence for a correspondingly appropriate type of integrated microfluidic biosensing [26].

#### *3.2. Acoustically Induced Cavitation Enables Rapid Micromixing*

E ffective mixing of fluids is important in several applications including chemical synthesis, biochemical reactions, and clinical diagnosis. As with any biochemical assay, for accurate colorimetric quantification su fficient mixing of fluids is important in order to homogenize the specimens with the corresponding reagents [27]. However, one of the limitations with microfluidics is the laminarity of flow in the device, which renders turbulent mixing di fficult. Here, by incorporating a piezo-actuated acoustic micromixer within the microfluidic design, we experimentally demonstrated the feasibility of controlled mixing of two non-reacting liquids through the inception of cavitation bubbles.

For convenience and ease of visualization, ethanol was stained blue and water was stained with red dye in all the experiments. Under continuous flow, the flow of liquids inside the microfluidic channel was extremely laminar (calculated Reynolds number ~5). The mixing ability of fluids under flow was tested in di fferent conditions, and no mixing was observed even for similar liquids of very high miscibility (blue ethanol colorless ethanol) in both with or without the actuation of the micromixer (Figure 3A(i)). For dissimilar liquids (water–ethanol), instantaneous di ffusion mixing was also negligible for both continuous flow and 'static' (no flow) conditions. Once the flow of liquid in the channel was stopped, actuation of the micromixer at 1000 Hz induced the formation of cavitation bubble in water (Figure 3A(ii)). We believe that PDMS substrates, being gas permeable facilitates the generation of bubbles, even for such low actuation frequencies, which in this case was useful for fluid mixing. Thereafter, reduction of the operating frequency to 800 Hz retarded the growth of the cavitation bubble, and at the same time creating micro-vortices in the fluids to initiate mixing. The real-time sequence of cavitation assisted fluid mixing is as shown in Figure 3B. Mixing of the fluids was instantaneous upon the inception of the cavitation bubble. Increasing the actuation frequency decreased the time taken for cavitation bubble inception (Figure 3C). However, with increased frequency, the growth of the cavitation bubble was also rapid and di fficult to control precisely. Therefore, to create steady cavitation, all micromixing experiments were conducted at an actuation frequency of 1000 Hz to initiate the cavitation bubble, and thereafter the operating frequency was reduced to 800 Hz to allow fluid mixing. The actuation frequency of our bulk acoustofluidic micromixer is around 100 fold lower than micromixing actuation using surface acoustic waves [27]. Lower actuation frequency could be useful for handling samples that contain living cells, as high frequency could induce undesired e ffects such as cell proliferation, di fferentiation, lysis etc. [28].

**Figure 3.** (**A**) (i) Image of the microfluidic channel with two inlets allowing laminar flow of two shades of ethanol (colorless and blue) without mixing when introduced from either of the inlets. (ii) Image showing the inception of cavitation bubble in water (red) to initiate the mixing of fluids (**B**) Image sequence showing the cavitation enabled mixing of water and ethanol inside the microfluidic channel. (**C**) Variation of the time taken for the inception of a visible cavitation bubble with respect to the piezo actuation frequency. Cavitation inception time was consistent (within 3% variation) for actuation frequencies greater than 1 kHz. ( **D**) variation in the fraction of the respective fluids mixed inside the microchannel with respect to time. To enable this fluid mixing, caviation was initiated at 1000 Hz actuation frequency, and the working frequency was reduced to 800 Hz, so as to retard the growth of the cavitation bubble.

The mixing efficiency was studied by measuring the concentration of red pixels in the section of the microfluidic channel initially containing ethanol (blue) and the concentration of blue pixels in the microchannel section initially containing water. Within 30 s, a mixing efficiency of more than 75% over a volume of ~3 μL of the fluid (Figure 3D). This is comparable to some of the previously published results. For example, Liu et al. [29] have reported cavitation assisted mixing of 22 μL of the fluids in ~105 s, and for the system proposed by [30], the mixing efficiency is governed by the flow rate which is 8 μL/min. Both these values are close to what we observed, thus corroborating with the efficiency achieved through our micromixing technique. Continued operation of the micromixer produces agglomeration of non-covalently bound particulates from the coloring dye which interfered with mixing. However, for proof-of-concept preliminary demonstration of fluid mixing, our experiments show that efficient mixing of small fluid sample volumes can be achieved by integrating the piezo-actuated acoustic micromixer that generates micro-vortices through the controlled induction of cavitation bubbles.

#### *3.3. Demonstration of On-Chip Optical Absorption*

One of the major drawbacks of absorbance-based detection in microfluidics is that it is considered to decrease in the optical path length with reduced usage of the sample volumes, which could directly affect the sensitivity of the system [31]. Therefore, in order to ascertain whether our device configuration was sensitive enough to support real-time optical absorption measurements, we conducted a preliminary non-calibrated colorimetric assay. For this proof-of-concept experiment, we demonstrated the detection of glucose [32] in cell culture media, based on its reaction with Benedict's reagent.

It is well known that glucose, a reducing sugar, induces the conversion of cupric ions (Cu<sup>2</sup>+) to cuprous form (Cu+) in the Benedict's reagen<sup>t</sup> [33]. This reaction brings about a change in color of the solution mixture, with the coloration being dependent upon the amount of glucose present. To firstly verify this principle, glucose-enriched media (hereafter referred to as glucose solution) was diluted to varying concentrations and mixed with the Benedict's reagent. The color changes in the mixture samples are as shown in Figure 4A. The color observed for the reactant mixture for undiluted glucose solution is indicated with the arrow. Upon heating the mixture, the color of those reactants changed to dark red as expected.

Before implementing this colorimetric reaction inside the integrated optical microfluidic setup, baseline optical absorption values at 635 nm wavelength were recorded by passing the glucose solution and the Benedict's reagen<sup>t</sup> individually through the microfluidic channel. Thereafter, to initiate the reaction, the liquids were pipetted simultaneously into the microfluidic chamber, and the piezoacoustic micromixer was actuated. To the naked eye, the liquids initially appeared colorless upon entering the micromixing chamber (Figure 4B at t = 0). However, with continued piezo-actuation, mixing of the fluids causes an increase in the turbidity of the reaction mixture (Figure 4B).

The liquids were allowed to mix for 60 s, after which the piezo-actuation frequency was increased to drive the liquids under supercavitation [22] through the optical detection zone and finally out of the channels. Upon image processing, the color of this reaction mixture inside the microfluidic chamber (Figure 4B) and the color of the ejected reaction mixture (indicated in Figure 4C) was observed to be very close to the externally prepared reaction mixture indicated in Figure 4A, The results for the measurement of the colorimetric optical absorption is presented in Figure 4D. As expected, at 635 nm wavelength, the optical absorption of the blue colored Benedict's reagen<sup>t</sup> was higher than the phenol-red tagged glucose solution (Figure 4D(i)). The initial randomness in the optical absorbance signal for the reaction mixture (Figure 4D(ii)), is we believe, noise due to the flow of cavitation bubbles in the reaction mixture. The mean optical absorption of the reaction mixture was significantly higher (*p* < 0.001) than the both individual liquids, presumably due to the increased turbidity obtained during the reaction (Figure 4E). The optical absorbance value of reactants mixed inside the microfluidic channels was very close (<5% variation) to that of the reaction mixture externally prepared by manually vortexing the reacting fluids, thereby further confirming the homogeneity of the micromixing process. The experiment was repeated 3 times, and the results were found to be consistent. Thus, the results

of this experiment demonstrates the implementation of optical absorption measurement, with our micromixing-enabled optical sensing setup.

**Figure 4.** Implementation of Benedict's reaction inside the optical microfluidic device for colorimetric detection of glucose in the presence of cell culture media. (**A**) Change in the color of the specimens for various concentrations of the glucose solutions, with the diluted samples showing shades of blue color. Reaction mixture upon heating turned to brick red color as expected. Arrow indicates the color obtained with undiluted glucose. (**B**) Variation, at different time points, in the turbidity of the reacting fluids (glucose solution and Benedict's solution) upon piezo actuated mixing inside the microfluidic channel. (**C**) Reaction mixture ejected out of the microfluidic channel, indicating the same coloration as indicated in panel-A. (**D**) Real-time absorbance measurements at 635 nm wavelength of (i) glucose solution and the Benedict's reagen<sup>t</sup> and (ii) Reaction mixture upon entering the optical detection zone. (**E**) Comparison of the optical absorption values of the working fluids. The reaction mixture shows a significantly higher absorbance (*p* < 0.001) than both the glucose solution and the Benedict's solution, due to the turbidity created upon mixing, thereby validating the colorimetric detection principle.

#### *3.4. Colorimetric Detection of Salivary Potassium*

With the colorimetric detection feasibility well established through the proof-of-concept glucose detection experiments, we further explored the possibility of implementing our microfluidics-integrated optical detection platform for clinically relevant real-time biosensing applications. Potassium monitoring and level correction is a routine but critical procedure carried out at the hospitals for patients undergoing cardiac, renal or neurological related diagnoses. Altered potassium levels in the body under hyperkalemia or hypokalemia may cause arrhythmia, abnormal blood pressure, cramping, twitching or paralysis of muscles leading to the development of abnormal cardiac rhythms [34,35]. Since whole blood assays currently adopted in hospitals for potassium monitoring is painful and sometimes unsuccessful, we reasoned that salivary diagnostics of potassium could be far less traumatic and could generate rapid results when performed in a point-of-care setting. To validate this hypothesis, we tested whole saliva samples through colorimetric optical absorption, for the measurement of absolute salivary potassium concentration.

The absorption measurements were conducted at 450 nm wavelength, as prescribed by the manufacturers of the potassium detection kit. Initial control experiments were conducted to measure the optical absorbance values of test samples with known potassium concentration. Unstimulated saliva samples were collected from random volunteers on the North CarolinaA&T State University campus. Samples were collected from 3 volunteers above the age of 65 years, 6 volunteers in the age group of 35–65 years, and 5 student volunteers under the age of 35 years. The results of our colorimetric potassium concentration measurement from the collected saliva samples are presented in Figure 5A. The overall estimated mean concentration of salivary potassium was found to be well within the range of typical salivary potassium levels as reported earlier [36]. The mean unstimulated potassium

concentration with people between the age of 13 and 35 has been reported to be ~21 mmol/L [37]. These observations match the data from our experimental results as well within 5% error, thereby validating our approach for a colorimetric optical absorbance-based salivary analysis method used herein.

However, contradicting observations have been made in literature regarding exact salivary potassium concentrations for patients with health conditions. We reasoned that there are several factors that could influence the salivary electrolyte concentrations, and a larger data set is required to make any robust comparison and inference using our device. However, the intent of the present work is far from delving into detailed biochemical analyses of saliva under different conditions. The objective of this work is to demonstrate the application of colorimetric sensing using microfluidic mixing for optical sensing. Therefore, in order to validate our optical measurements, we measured the optical absorbance of three random saliva samples treated with the chromogenic agents on-chip, and compared the results to the optical absorption of those samples measured externally using a standard spectrophotometer (Cary 6000i UV/Vis/NIR, Agilent, Santa Clara, CA, USA). The results shown in Figure 5B demonstrate no significant difference between the absorbance values measured through these two techniques, thereby validating our on-chip analysis approach. The time-dependent on-chip absorbance measurements and the absorbance spectrum obtained using the spectrophotometer for one of the samples is presented in Supplementary Figure S1A.

**Figure 5.** Results of the colorimetric optical absorbance detection of potassium from unstimulated whole saliva. (**A**) Measurement of estimated potassium concentration among people (15 volunteers) of different age groups indicating that older people (>65 y age) show a significantly lesser (*p* < 0.01) salivary potassium content than people in the age group of 35–65; and also significantly reduced potassium than people in 20–35 age group. In general people above the age of 35 show significantly lesser (*p* < 0.05) potassium than people under the age of 35. (**B**) Comparison of the optical absorbance values of salivary samples mixed with chromogenic agen<sup>t</sup> measured on chip and off-chip (using an ultraviolet/visible (UV/Vis) spectrometer) shows no significant difference (*p* = 0.2643 with paired t-test) for the samples analyzed through the two measurement techniques.

Thus, the results of our work show that the colorimetric optical absorption based biosensing principle with integrated micromixing can be readily applied to the lab-on-a-chip format for real-time analysis of metabolites from saliva to generate diagnostic results with a quick turn-around.
