Deciphering Flow Cytometry Electronics

Transcription

1 Deciphering Flow Cytometry Electronics Fan Xiong Life Science Group Bio-Rad Laboratories Hercules, CA, USA Murat M. Tanik Electrical and Computer Engineering Department University of Alabama at Birmingham Birmingham, AL USA Abstract This paper introduces flow cytometry electronics to the Electrical Engineering community and briefly explains critical electronics aspects which confuse biologists. Furthermore, this paper explains flow cytometry from an interdisciplinary point of view and points out open research opportunities in electronics. Keywords flow cytometry; cell sorting; digital signal processing; ADC; baseline restoration I. INTRODUCTION Flow cytometry is the process of measuring the physical or biochemical characteristics of cells or cell-like particles while they pass within a fluidics stream intersected by a laser beam. This intersection is known as the interrogation point or point of analysis. Flow cytometry has been widely used in DNA analysis, immunophenotyping, and cell sorting. Flow cytometry is interdisciplinary, requiring knowledge of biology, biochemistry, physics, optics, mechanics, electronics, statistics, and data analysis. Based on a market report, the global flow cytometry market is expected to reach $3.7 billion by 218, at a compound annual growth rate of 1% from 213 to 218. The number of publications in this field is growing at a rate of more than 1, papers per year since 21, see Fig.1. However, among the large pool of flow cytometry publications, there are very few publications covering electronics [1-1]. J. Paul Robinson wrote a 215 New Year message to the flow cytometry community themed wake up cytometry field covering the current shortcomings in the field [11]. The purpose of this paper is to explain the flow cytometry electronics to scientists in general and to biologists in particular. We also aim to raise awareness on this research topic especially in the Electrical Engineering community. There are four major parts of flow cytometry: optics, fluidics, electronics, and data analysis. Fig.2. shows a typical flow cytometry cell sorting system. The upper half shows a flow cytometer system and the lower half shows the sorting system. The optics system contains lasers, optical filters and other optical components to make sure that the laser is aligned well with the stream and to guide the resulting fluorescence and/or laser scattering signal to the appropriate optical detector. The fluidics system generates and maintains a stable stream of fluidic, known as the sheath stream. Sample cells are introduced into the center of this sheath stream and remain centered as they pass through the interrogation point. The resulting optical signal, whether scatter or fluorescent, is directed to the appropriate detector by an optical path. At that point a photomultiplier tube (PMT) converts the optical signal into an electrical one, and then the electronics system processes the signals and extracts information like pulse height, area and width. Data analysis is usually done by software which displays histograms or density plots of cell statistics. At this point, we have described a flow cytometer, which is only an analytical instrument. In addition, cell sorting can be done based on the analytic information that s been acquired. In this case, the electronics need to decide if the cell that s been analyzed is of interest, and if it is, apply a charge to it so it can later be electrostatically deflected into a collecting tube. History of Flow Cytometry and qpcr Publications Number of Publications Flow Cytometry qpcr Year Fig.1. Flow cytometry and Real-time polymerase chain reaction (qpcr) publication statistics

2 more like a black box, so this paper will be focused on electronics. This paper is organized as follows; Section II talks about electronics with a focus on Analog to Digital Convertor (ADC) and signal processing, while Section III discusses and concludes the paper. Fig.2. Flow cytometry cell sorting system [12] Flow cytometry system is complicated and highly interdisciplinary. A typical flow cytometer design team has optical, mechanical, and electrical engineers along with biologists and programmers. A flow cytometer will not work if any of the five components mentioned above are not working. From an interdisciplinary point of view, understanding PMTs requires a good physics background; optical noise affects instrument sensitivity which could in turn affect the ability to identify dim cells. Designing an optical system with multiple scatter and fluorescence paths requires expertise in optics and collaboration with mechanical engineers for precision laser stream alignment. Generating and maintaining stable droplets, and designing fluidics manifolds and tubing requires mechanical expertise. How the cells are distributed in the droplets could be studied with statistics like Poisson distribution. From the application point of view, the users are mainly biologists, and an experienced user also has good knowledge of optics and data analysis while electronics are II. ELECTRONICS This section will not cover all the parts of flow cytometry electronics, but briefly describes ADCs and digital pulse processing, since these two parts are perplex to the biologists most. A. Analog or Digital Electronics When Shapiro wrote his book Practical Flow Cytometry in 1998, the electronics were generally analog. There have been a great number of discussions on analog or digital electronics in the past. The electronics have evolved from analog to semi-digital, then to all digital electronics. Fig.3. shows the block diagram of the three different electronic systems. The advantage of the digital electronics is easier triggering on parameters, accurate log conversions, and the full compensation matrix implemented with digital electronics. Modern digital signal processing could also be easily applied. The major components of the electronics system are the detectors, which are usually photo diodes (PMT), preamp circuit, baseline restoration circuit, compensation circuit, log conversion, integrator/peak detection circuit, and ADC. Christopher Snow wrote a review paper on flow cytometer electronics which has a good coverage of each of these components mentioned here [7]. Below the relationship of ADC to dynamic range, along with a more vivid description of the digital signal processing is given. B. The ADC The dynamic range of flow cytometry data largely depends on the ADC speed and resolution. Let s take a look at the resolution first. It is difficult to say how wide of a dynamic range is sufficient, but the basic need is 4 decade log display since the positive population of some brightly stained cell could be 1, times brighter than the negative population. In this case, an ADC with a signal to noise ratio (SNR) of at least 8 db is needed. At this point, given the speed requirements, 14/15/16 bits ADCs are the popular ADCs used for commercial cell sorters.

3 Traditional Flow Cytometer PreAmp Compensation circuit Log Amp Integrator/ Peak ADC Semi-Digital Flow Cytometer Integrator/ PreAmp ADC Peak Compensation Log Conversion Digital electronics All Digital Flow cytometer PreAmp ADC Triggering, baseline restoration Integrator/peak detection Compensation Digital electronics Log Conversion Fig.3. Flow cytometer electronics Fig.4. Data acquisition and processing system If we convert the 4 decade dynamic range to SNR, which is SNR= 2log(1^4)=8. For an ideal 14 bit ADC, the SNR could be estimated as SNR = 6.2* = Let s take the Becton Dickinson (BD) FACSDiVa electronics for an example. The ADC is a 14 bit ADC AD924AS from Analog Device. The datasheet shows a 78.5dB SNR and so the effective number of bits could be calculated as: ENOB = ( )/6.2=12.75 bits, which can be converted to 2^12.75= 6888 bins. This value shows the maximum dynamic range the ADC is capable of resolving. A typical flow cytometry input voltage range is 1mv ~ 1v. Table 1 shows the relationship of the input voltage and the discrete value based on this ADC over a 4 decade range. TABLE I. Voltage INPUT VOLTAGE AND NUMBER OF BINS 1mv 6 1mv 68 1v 688 1v 6888 Number of bins It can be seen from Table I that there are two issues here: 1) the dynamic range is not 4 decade yet since the upper decade is 6888 bins; 2) the lower end resolution is very low, and it is very likely that signal is buried in noise since the bottom decade only has 6 total bins. Digital signal processing techniques like filtering can improve the SNR to above 8dB. For example BD Aria 1MHz 14-bit ADC conversion data is smoothed to produce 18-bit data; The Beckman Coulter (BC)

4 EPICS XL 4M 15 bit ADC conversion data is smoothed to produce 2-bit data. But this still doesn t resolve the lower end resolution issue. Measuring pulse area rather than height can help this problem to some extent since integration of area under the peak effectively filters out high frequency noise. Let s take a look at ADC speed in terms of speed. For the Analog Device ADC AD924AS, the sampling speed is 1MHz. A typical pulse width is in the range of 1us ~ 1us. Let s take 3us for example, there are 3us pulse/ (1/1M) = 3 samples per pulse. Pulse height: bits Pulse area: log2(3) = 17 bits The log2(3) would be the improvement in dynamic range given by averaging a steady signal. In reality, we don t have a steady signal but more of a Gaussian shaped signal. Understanding the exact improvement given a Gaussian shaped signal would be an interesting research activity. C. Digital pulse signal processing The digital pulse signal processing for flow cytometry signal is nothing more than a typical DSP system described in the book [13]. Fig.4. shows how the system applies to flow cytometry area. The raw PMT signal is filtered, then digitized, SNR is improved after signal processing, and finally the extracted information is displayed in the format of histograms or density plots. Fig.4 also shows a raw PMT signal and the density plot. Common digital processing components are filtering, baseline restoration, pulse peak detection and area integration. There are three types of light detection signals: forward scatter, side scatter and fluorescence. Fig.5. conceptually shows how the three types of signals look like. Usually the raw signal is quite noisy as shown in Fig.6. A cleaner signal after filtering some of the noise is shown in Fig.7. The digital filter used here could be an average filter, FIR filter or even more complicated filters. A discrete signal example after ADC is shown in Fig.8. Fig.6. Raw PMT signal Fig.7. Filtered PMT signal Fig.8. Digitized waveform signal Baseline restoration is not covered much in the existing literature; the basic idea is to subtract the constant background level from each digital value. Fig.9. shows the concept and simulation results. Fig.5. Forward scatter, side scatter and fluorescence signals

5 Fig.9. Baseline restoration Pulse signal extraction is more obvious. There are usually three parameters extracted: height, area and width as shown in Fig.1. Pulse height is the maximum value of the digitized signal. Pulse area is the sum of the digital values over the waveform profile. Pulse width is a little bit different, which could be the full width above the threshold or the width where crosses ½ height level. Fig. 1 Pulse height, area and width extraction III. DISCUSSION AND CONCLUSION Flow cytometry field is highly interdisciplinary. Even the subarea of digital pulse signal processing is cross disciplinary. For example, there are a wide range of papers in nuclear engineering field which utilizes the same techniques [14]. There are many open research opportunities in the flow cytometry electronics area compared to so few research groups and publications. There are even fewer published works on the charging and sorting system. To name a few research opportunities here based on Section II: 1) To achieve 5 decade or even higher dynamic range. How to improve SNR with available commercial ADCs? How to get better SNR by exploring different filtering techniques? 2) Information extraction. Are area, height, and width enough? How much weight should be put to area or height? 3) How to optimize high speed cell sorting. 4) Another research opportunity on the electrostatics side is to understand the relationship between consecutively charged droplets. It has been observed that the amount of deflection of a droplet is dependent on the time history of the charge state of preceding droplets. The process of compensating for this is known as defanning. 5) On the other side, flow cytometers are expensive. How to reduce the cost could be another research direction. Shapiro, mentioned above, is actively working on low cost flow cytometers. More inputs from the IEEE community would greatly help to advance the field. IV. ACKNOWLODGEMENT We are greatly thankful to Ed Marquette, Frank Shen, Carol Oxford and Mike Kissner from Bio-Rad Laboratories for their review and discussions on this paper. REFERENCES [1] H.M Shapiro, Practical Flow Cytometry, Wiley, 1994 [2] H.M Shapiro, Practical Flow Cytometry, John Wiley & Sons, 25 [3] H.M Shapiro, NG Perlmutter, PG Stein, A flow cytometer designed for fluorescence calibration, Cytometry, vol. 33, no.2, pp , Octber [4] van den Engh G, W Stokdijk, Parallel processing data acquisition system for multilaser flow cytometry and cell sorting, Cytometry, vol. 1, no. 2, pp , May [5] P.M.A. Sloot, P. Tensen, and C.G. Figdor, Spectral Analysis of Flow CytometricData: Design of a Special Purpose Low-Pass Digital Filter, Cytometry, vol.8, pp. 545, [6] NA Zilmer, M Godavarti, JJ Rodriguez, TA Yopp, GM Lambert, DW Galbraith. Flow cytometric analysis using digital signal processing, Cytometry, vol.2, no.2, pp , Jun [7] C. Snow, Flow cytometer electronics, Cytometry A, vol. 57, no.2, pp , Feb. 24. [8] S Murthi, S Sankaranarayanan, B Xia, GM Lambert, JJ Rodríguez, DW Galbraith, Performance analysis of a dual-buffer architecture for digital flow cytometry, Cytometry A, vol.66, no.2, pp , Aug. 25. [9] MA Naivar, JD Parson, ME Wilder, RC Habbersett, BS Edwards, L Sklar, JP Nolan, SW Graves, JC Martin, JH Jett, JP Freyer JP, Open, reconfigurable cytometric acquisition system: ORCAS. Cytometry, vol. 71A, no. 11, pp , 27. [1] F.R.E. De Bisschop, "Electronic Gating for Particle/Cell Counting and Sizing, DSP-operated," Instrumentation and Measurement, IEEE Transactions on, vol.58, no.9, pp , Sept. 29. [11] January/thread.html#start [12] C. Oxford, K. Kroeger, M. Ma, A. Vandergaw, D. Fox, S. Hunter, prodrop technology a novel approach to automated drop delay measurement and monitoring, 28th Congress of the International Society for Advancement of Cytometry (Cyto 213). [13] S. W. Smith, The Scientist and Engineer's Guide to Digital Signal Processing, California Technical Pub., 1997 [14] G. F. Knoll, Radiation Detection and Measurement, 4 th edition, John Wiley & Sons, 21.