Smartphone-Imaged HIV-1 Reverse-Transcription Loop-Mediated Isothermal Amplification (RT-LAMP) on a Chip from Whole Blood

Viral load measurements are an essential tool for the long-term clinical care of hum an immunodeficiency virus (HIV)-positive individuals. The gold standards in viral load instrumentation, however, are still too limited by their size, cost, and sophisticated operation for these measurements to be ubiquitous in remote settings with poor healthcare infrastructure, including parts of the world that are disproportionately affected by HIV infection. The challenge of developing a point-of-care platform capable of making viral load more accessible has been frequently approached but no solution has yet emerged that meets the practical requirements of low cost, portability, and ease-of-use. In this paper, we perform reverse-transcription loop-mediated isothermal amplification (RT-LAMP) on minimally processed HIV-spiked whole blood samples with a microfluidic and silicon microchip platform, and perform fluorescence measurements with a consumer smartphone. Our integrated assay shows amplification from as few as three viruses in a ~ 60 nL RT-LAMP droplet, corresponding to a whole blood concentration of 670 viruses per µL of whole blood. The technology contains greater power in a digital RT-LAMP approach that could be scaled up for the determination of viral load from a finger prick of blood in the clinical care of HIV-positive individuals. We demonstrate that all aspects of this viral load approach, from a drop of blood to imaging the RT-LAMP reaction, are compatible with lab-on-a-chip components and mobile instrumentation.

lysis buffer, RT-LAMP was performed in the thermocycler with identical mastermixes containing reaction buffers, enzymes, and primers, but where the sample portion consisted of various ratios by volume of whole blood to lysis buffer: 1: 4, 1: 2, 1:1, or 1: 0. Viruses were spiked into lysed blood following the mixing in order to keep the virus concentration identical between samples. Results. Figure S2(a) shows threshold time with the 1: 0 case omitted (since it did not exhibit amplification). The results showed differences in threshold time compared to 1 : 4 of 0.59% and 1.03% for 1: 2 and 1:1, respectively, at 670 vp· RXN -1 and a difference in threshold time compared to 1: 4 of 0.97% and 3.09% for 1: 2 and 1:1, respectively, at 67 000 vp· RXN -1 . A standard t test produced P compared to 1: 4 for 1: 2 and 1:1, respectively, of 0.9247 and 0.8444 at 670 vp· RXN -1 , and 0.1604 and 0.0138 at 67 000 vp· RXN -1 . Figure S2(b) compares the average maximum baselinesubtracted fl uorescence intensity of each condition, demonstrating the trend of increased quenching of fl uorescence as absolute blood content increases. In this case, the bar for the 1: 0 (no lysis) condition merely represents the fluctuation of noise, as no amplifi cation was observed in these samples.
Following this characterization, tests were performed to determine if samples prepared at 1 : 2 or 1 :1 ratios of blood to lysis buffer could be adequately imaged on-chip with the fluorescence microscope imaging apparatus. This process proved problematic, however, and subsequent lysed blood measurements were performed at the 1: 4 ratio. Discussion. The characterization of blood-to-lysis buffer ratios confirms that the reduction of fluorescence signal seen in lysed blood versus purified RNA is related to a blood component and not to a component of the lysis buffer, and demonstrates a trend of increasing fluorescence quenching Figure S1. (a) Cross-section of pyramidal well with dimensions indicated; (b) mold of pyramidal well with dimensions indicated.

Cell lysis buffer ratio
Methods. For measurements comparing the ratio of blood to

Research
Engineering Volume 1 · Issue 3 · September 2015 www.engineering.org.cn with increasing blood content. Initially, it was thought that decreasing the volume of lysis buffer relative to the whole blood sample might allow for larger volumes of sample to be employed in the 25 μL or 60 nL reactions, and therefore might increase the detection limit of our platform with respect to the viremia of the original sample. Deviating from the lysis buffer ratio described by Curtis et al. [17], however, proved to be problematic on-chip due to inadequate fl uorescence intensity in the microscopy imaging; therefore, we decided to maintain the 1: 4 ratio for subsequent experiments.
The observation that blood concentration affects the fl uorescence intensity suggests that one or more components of the lysed blood have a diminishing effect but do not completely quench the fluorescence signal except for when no lysis is performed, while the lack of effect on threshold time suggests that blood components do not interfere with reaction kinetics. This is a critical result in order for quantifi cation of the sample based on reaction kinetics to be possible.

Betaine concentration in the RT-LAMP reaction
Initial experiments were performed with 0.8 mol· L -1 betaine based on the concentration used by Curtis et al. [17]. However, at a stock concentration of 5 mol· L -1 , the betaine reagent occupied 4 μL of the 25 μL reaction-more than any other single reagent. In order to include a greater volume of sample in each reaction, we cut the betaine concentration to 0.4 mol· L -1 , a reaction concentration also seen in Ref. [41]. We performed a control experiment to determine if the change in betaine concentration would have any signifi cant effect on the lower limit of detection. Figure S3 shows the raw fl uorescence curves from this test, which was performed with low concentrations of purified viral RNA in water. In this test, faster threshold times are observed in 0.4 mol· L -1 betaine compared to 0.8 mol· L -1 betaine. For the smaller concentration (an average of four viruses per reaction), a third of the reactions amplifi ed in the 0.4 mol· L -1 betaine condition while two thirds of the reactions amplifi ed in the 0.8 mol· L -1 condition. Although this is not a thorough characterization of the effects of betaine, we considered it adequate indication that the effect of betaine is not signifi cant.

Lysed blood microchip LAMP with microscope imaging
A microchip RT-LAMP experiment with lysed whole blood spiked with viral RNA was performed and imaged with the fluorescence microscope as described in the main text. Figure S4 shows the fl uorescence curves and threshold time curve. This experiment was an intermediate step between microchip LAMP with purifi ed RNA in water imaged with a microscope and RNA-spiked lysed whole blood imaged with the smartphone. The data are provided here for thoroughness.

Lyophilization of RT-LAMP reagents
To determine whether RT-LAMP reagents could be prepared with a freeze-drying method, we prepared an RT-LAMP mastermix containing buffers, enzymes, primers, and intercalating dye. The mastermix was aliquoted into 0.2 mL reac- www.engineering.org.cn Volume 1 · Issue 3 · September 2015 Engineering tion tubes and frozen at -20 °C overnight. Four of the frozen reactions were left in the freezer, while eight were kept on ice and transferred to a 2 L Labconco Freeze Dry System. The samples were left in the freeze drier for 5 min after the chamber reached full vacuum, after which the system was vented and the samples were kept slightly below ambient temperature for several hours. Fresh RT-LAMP reactions were then prepared and compared to the frozen mastermixes (kept at -20 °C throughout) and the freeze-dried reagents. Figure S5 depicts the results. Of eight freeze-dried mastermixes, one did not amplify while another showed delayed amplifi cation. Six of the eight reactions amplifi ed in a reasonable time, although threshold time was delayed and less consistent compared to fresh or frozen mastermixes.
We consider this a promising initial test demonstrating the feasibility of freeze-drying RT-LAMP reagents. This protocol was not compatible with on-chip reactions, in which 60 nL droplets evaporated rapidly when exposed to air.

Statistical model for digital droplet LAMP
The distribution of viruses in small droplets as performed in our method and in any digital PCR or LAMP process is a binomial distribution in which the probability of a "success" is defi ned by Poisson statistics [36,42,43]. The virus concentration from the number of positive droplets in the assay can therefore be calculated using Eq. (S1) [36,44]: where N is the total number of droplets; λ is the average number of viruses in a droplet; and x is the number of positive droplets. The concentration of viruses, therefore, is determined by dividing both sides of Eq. (S1) by the volume of the sample, v.
The uncertainty of this measurement can be determined by calculating a 95% confidence interval (CI) (α = 1.96), the classic form of which is given here [39,44]: where p is the proportion of PCR reaction that amplifi ed (x/N). However, Shen et al. employed a more sophisticated calculation of confi dence intervals based on the so-called "Wilson" method [39]: To illustrate this method, we will consider a practical fi nger prick test scenario, where a 10 μL sample is obtained and we assume that 90% of the sample is distributed into our RT-LAMP droplets (60 nL total volume, of which 4.8 nL comes from blood), while 10% is lost to dead volume in the microfl uidics. Examining low values for p demonstrates the increasing uncertainty of the digital method as extremes are approached. The 95% confi dence interval calculated by the Wilson method indicates an upper limit of 426 mL -1 in the case that no droplets amplify in a valid test. One positive droplet of 1875 corresponds to viral loads in the range 20-629 mL -1 .
While digital LAMP from a finger prick of blood cannot rival state-of-the-art tests (which have a lower limit of detection of 10 mL -1 in plasma and of about 5 mL -1 in whole blood), it does offer a working range that would be useful in remote settings where no alternative viral load measurement method is available.