Calibration of the accuracy and intrinsic noise of mass measurements
To fine-tune the absolute mass calibrations and to reveal the intrinsic noise of our methods, we first made a series of measurements on red blood cells (chicken and human). These cells are commonly used cytometry standards, in large part because they have undergone terminal differentiation, so state differences are minimized and this leads to a more nearly homogeneous sample. The standard deviations of our measurements these RBCs should be more nearly limited by the technical noise of our method.
We prepared fresh blood smears from a human subject on quartz and mounted in 50% glycerol, then collected images at 280, 260, and 220 nm. For intracellular protein in mammalian somatic cells we developed the 220 nm/260 nm algorithm as a low noise method that benefits from the high molar extinction of all proteins at 220 nm. We had selected human red blood cells as a validation marker for deep UV imaging because their dry mass is comprised of 95% hemoglobin (12). An effective extinction coefficient for protein can be calculated as a weighted sum of the epsilons for hemoglobin (ε-280 nm = 118,872 M−1 cm−1; ε-260 nm = 116,376 M−1 cm−1) and an "average" protein (ε-280 nm = 54,129 M−1 cm−1; ε-260 nm = 36,057 M−1 cm−1). For the specialized case of hRBC, total protein is dominated by hemoglobin: εeffective 280 nm = 115,634 M−1 cm−1 and εeffective 260 nm = 112,360 M−1 cm−1.
A representative protein mass map is shown in Figure 2. The protein map shows a narrow distribution of nearly uniform discs. Using the protein map to generate an automated segmentation mask, we obtained a total protein mass of 26.6 pg, assuming a protein distribution of 95% hemoglobin (Fig. 2B, mean ± s.d., 27.0 ± 8.0 pg).
Figure 2. Human red blood cells (hRBCs). A: Protein (substantially hemoglobin) mass map. Histograms (n = 1034) of (B) total protein mass by 260 nm/280 nm absorption (see text), and (C) total protein mass by 220 nm/260 nm absorption.
Download figure to PowerPoint
These numbers are consistent with the hemoglobin quantities (10–30 pg) measured using classical hematology methods and by bulk protein determinations in lysates [i.e. by lysing RBCs, reacting hemoglobin with cyanide, and measuring absorption at 540 nm (15)]. The CV of the protein histogram is 0.30, which provides an upper bound on the noise for the mass mapping technique when using the classical 260 nm/280 nm wavelength pair. This is comparable to good imaging cytometry with fluorescent markers—a respectable result, particularly since mass mapping provides absolute quantitative measurement and is based on native contrast.
Some further improvement in the measurement is possible through the new 220 nm/260 nm data reduction. Although we could not find good literature values for the molar extinction coefficient of hemoglobin at 220 nm (ε-220), we derived this number, as previously (8) using our measurements with the 260 nm/280 nm literature values as calibration and requiring overlay of the 260 nm/280 nm and 220 nm/260 nm histograms. This resulted in an empirical ε-220 nm for protein of 600,000 M−1 cm−1. We cross-checked this empirically derived epsilon by measuring the UV absorption spectra for hemoglobin solutions (1.0 mg/ml, 0.1 mg/ml, 0.01 mg/ml, fused-silica cuvette, 1-cm path length), yielding an εeffective 220nm of 50,5620 M−1 cm−1. However, RBCs in solution showed significant differences in absorption at deep UV wavelengths, compared to hemoglobin solutions (16). Interpolation of RBC absorption at 220 nm and 260 nm yielded an εeffective 220nm of 606,000 M−1 cm−1.
Using the now-calibrated 220 nm/260 nm method, we calculated peak protein content at 26.4 pg (Fig. 2C), with the CV contracted to 0.17 (mean ± s.d., 27.5 ± 4.6 pg). This is a reduction by a factor of nearly two in standard deviation compared to the 260 nm/280 nm method.
We next addressed calibration to a known nucleic acid standard using chicken RBC nuclei, a common calibration standard for genomic content in flow cytometry (17–20). As an approximation, we assumed the same effective epsilons for protein as for hRBCs.
Our results show that whole cRBCs (nuclei shown in Fig. 3A) contain peak nucleic acid (Fig. 3B) content of 8.5 pg (n = 376 cRBCs; mean ± s.d., 8.5 ± 1.6 pg) and peak protein (Fig. 3C) content of 24.0 pg (mean ± s.d., 27.0 ± 6.1 pg). The CVs are nearly as good as the simpler case of hRBCs (Fig. 2C), for both nucleic acid (CV= 0.18) and protein (CV = 0.23). For the nuclear compartment we found a nucleic acid peak of 3.00 pg (mean ± s.d., 3.2 ± 0.8 pg), from 468 nuclei (Fig. 3D). This represents the total nucleic acid, RNA and DNA, in the cRBC nucleus.
Figure 3. Distribution of nucleic acid and protein mass for chicken red blood cells. A: Nucleic acid mass map in extracted, RNase-treated cRBC nuclei. Histograms showing distribution of total nucleic acid (B) and protein (C) mass for 376 whole cRBCs. D: Histogram of nucleic acid in isolated nuclei (n = 468). E: Histogram of DNA mass from RNase-treated cRBC nuclei (n = 1170).
Download figure to PowerPoint
To directly measure the genomic DNA content of cRBCs, we treated extracted cRBC nuclei with RNase and isolated nuclei automated segmentation (Figs. 3A and 3E). The extracted nuclei showed a peak DNA mass of 2.30 pg (n = 1170 nuclei, mean ± s.d., 2.2 ± 0.4 pg). In these same nuclei, the CV in Hoechst fluorescence is respectable but appreciably higher (0.38) compared to the NA measurement by UV (Fig. 3E, CV = 0.18): that is we were able to achieve better signal-to-noise ratio with the UV quantitation method than wide field fluorescence. Our measurement for a White Bovan male falls at the low end of the range of the literature values for various other breeds of male and female chickens (2.33–2.54 pg, Ref.18–25).
By subtracting mean genomic DNA (2.30 pg) from the mean nucleus-isolated total NA (3.00 pg), we estimated nuclear RNA to be ∼0.70 pg for cRBCs. Thus in cRBC, the nucleus has markedly higher DNA than RNA, with the ratio typically being closer to 1:1 for somatic mammalian nuclei (see text for Fig. 5, Ref.3). In addition, the whole cell (total) nucleic acid was measured to be 8.5 pg, and with 3.0 pg (total NA) in the nucleus, this leaves ∼5.5 pg (total NA) localized to the cRBC cytoplasm.
An important broader result is that the calibrations in RBCs have proven that the mass mapping method can determine population distributions and compartmental localization of NA and protein with good overlay to previous hematology methods. Furthermore, in populations of 500–1000 cells we achieve CV values of 0.15–0.23, which is equivalent or better than imaging cytometry using indirectly calibrated fluorescence.
Whole cell protein and nucleic acid
We then proceeded to build histogram distributions of five cell lines under conditions of exponential expansion. Figure 4 shows the distributions for whole cell nucleic acid and protein in CHOK1, Jurkat, QGY-7703, S3T3 and Mel-10 cell lines. We also extended the population analysis to mouse embryos extracted at the eight-cell stage (50 hours post-fertilization), shown in the same figure.
Figure 4. Histograms of whole cell nucleic acid and protein masses. Histograms for six cell types are shown (CHO, Jurkat, QGY-7703, Swiss3T3, Melanoma-10, and mouse embryo) for nucleic acid (top) and protein (bottom). All mass maps were calculated using 220 nm/260 nm image pairs. Manual segmentation was used to delineate regions of interest (ROIs) for individual cells. Histograms were fitted using the ‘polyfit’ function in MatLab. The peak was found for that fitted curve; mean and standard deviation were calculated across all objects of a given cell type.
Download figure to PowerPoint
Our first objective was to provide absolute anchor points for fluorescence-based cytometry. Generally, the peaks of the fitted curves (MatLab “polyfit”) lie close to the mean of the distributions. The cells can be ranked (Table 1) in order of most probable nucleic acid content (G1 peak): Jurkat (17.2 pg), CHO (31.6 pg), QGY-7703 (39.6 pg), S3T3 (73.1 pg), Mel-10 (31.6 pg). If we assume that the genomic DNA component of this total is near the 2C diploid cell at G1, then the whole-cell RNA content is 3X to 15X the genomic content. The total whole-cell protein distributions have qualitative similarities to total nucleic acid, (Fig. 4). The same ordering of most probable mass in the G1 peak is obtained for the cell lines with total protein values: Jurkat (32.1 pg), CHO (82.4 pg), QGY-7703 (251 pg), S3T3 (207.7 pg), and Mel-10 (182.6 pg).
Table 1. Nucleic acid and protein masses for cell lines (ranked by NA)
| || n||Size (μm2)||Nucleic acid (pg)||Protein (pg)|
|Peak||Mean ± s.d.||Peak||Mean ± s.d.|
| Jurkat||466||74.1 ± 23.6||17.2||20.0 ± 7.1||32.1||39.4 ± 15.7|
| CHO||368||216 ± 106||26.0||28.2 ± 9.2||82.4||90.6 ± 36.4|
| QGY-7703||489||638 ± 254||39.6||44.1 ± 16.7||251||303 ± 93.2|
| Swiss 3T3||441||690 ± 515||73.1||89.8 ± 45.9||207.7||305.2 ± 198.6|
| Melanoma||337||665 ± 679||31.6||92.4 ± 97.6||182.6||336.6 ± 300.2|
| Jurkat||493||44 ± 25||9.3||11.3 ± 4.5||20.9||23.6 ± 10.7|
| Melanoma||269||104 ± 50||15.6||16.4 ± 7.3||66.2||68.5 ± 36.6|
| CHO||325||57 ± 29||12.6||17.3 ± 11.6||37.6||44.8 ± 31.2|
| QGY-7703||509||116 ± 45||14.8||18.2 ± 4.7||99.3||101 ± 33.8|
| Swiss 3T3||306||165 ± 59||28.3||37.8 ± 15.8||79.9||97.6 ± 46.8|
Mass in isolated nuclei
We then used the overlay of Hoechst stain to automate a segmentation and separate mass components in the nuclear and cytoplasmic compartments. In general we found less than half the total nucleic acid in the nucleus (Fig. 5). Arranged in ascending most probable (peak) mass (Table 1), the sequence is Jurkat (9.3 pg), CHO (12.6 pg), QGY-7703 (14.8 pg), Mel-10 (15.6 pg), and S3T3 (28.3 pg). In this case the S3T3 line is highly exceptional, with a much greater total NA in the nucleus – even when compared to the melanoma (Mel-10) line or the mouse blastomere (23.2 pg, Table 2). It must be quickly added that the Mel-10 line shows a highly distorted NA distribution in the nucleus with both a low-mass (7.4%, 20/269 cells) and a high-mass (14.1%, 38/269 cells) shoulder. The high-masses are easily assigned to the cells at high ploidy number, however identity of the low-mass shoulder population is less obvious. If the peak values in the histograms are assigned to the G1 phase with nominal 2C genomic complement, then CHO, Jurkat, QGY-7703 and Mel-10 cells all would have a RNA:DNA ratio of between 1:1 or 1.5:1 in the nucleus. The S3T3 line would appear to have more than twice as much RNA as DNA, and the mouse blastomere, at the eight-cell stage, has a nuclear RNA:DNA ratio of nearly 4:1.
Figure 5. Histograms for nucleus-isolated, nucleic acid, and protein masses. Histograms for six cell types are shown (CHOK1, Jurkat E6, QGY-7703, Swiss3T3, Melanoma-10, and mouse embryo) for nucleic acid (top) and protein (bottom), in the nucleus. All mass maps were calculated using 220 nmn/260 nm absorption images. Corresponding Hoechst fluorescence images were used as masks in an automated segmentation algorithm. Histograms were fitted using the ‘polyfit’ function in MatLab. The peak was found for that fitted curve; mean and standard deviation were calculated across all objects of a given cell type.
Download figure to PowerPoint
Table 2. Nucleic acid and protein masses, in picograms, for primary cells
| || n||Nucleic acid (pg)||Protein (pg)|
|Peak||Mean ± s.d.||Peak||Mean ± s.d.|
| Chicken RBC||376||8.5||8.5 ± 1.6||24.0||27.0 ± 6.1|
| Human RBC||1034||−0.4||−0.2 ± 0.5||26.4||27.5 ± 4.6|
| Mouse embryo||54||104.0||104.3 ± 22.8||1267.8||1409.9 ± 329.2|
| Chicken RBC||468||3.0||3.2 ± 0.8||3.0||5.3 ± 3.8|
| Mouse embryo||54||23.2||23.2 ± 3.2||247.3||253.3 ± 67.7|
| Mouse embryo||54||1.7||1.8 ± 0.4||17.2||19.9 ± 6.4|
| Chicken RBC||1170||2.3||2.2 ± 0.4||3.8||3.8 ± 1.2|
When we compared the nucleus to cytoplasmic NA, the nucleus contains approximately half of the whole cell total. We confirmed this when we plotted cytoplasmic and nuclear quantities across CHO, QGY-7703, Jurkat, and Mel-10 cell types (n = 1420 matched nucleus-cytoplasm pairs; R = 0.50; slope = 1.05). In contrast the eight-cell mouse blastomere sequesters only one quarter of its total NA in the nucleus. Note that the there is a tight CV (0.14) for the total NA content in the nuclear compartment of the blastomeres, even when compared across multiple embryos (Table 2).
Figure 5 also shows the distributions of protein in the nuclear compartment. For CHO, QGY-7703, and S3T3 fully 40%–45% of the total cell protein is found in the nucleus. This increases to more than 65% for Jurkat (a lymphocyte) and, as would be expected, falls to 20% of the total protein in the (much larger) blastomere. We observed the following sequence, in order of increasing peak protein mass in the nucleus: Jurkat (20.9 pg), CHO (37.6 pg), Mel-10 (66.2 pg), S3T3 (79.9 pg), QGY-7703 (99.3 pg), and mouse embryos (247.3 pg).
For most cells that we measured there is ∼3X more protein than nucleic acid in the whole cell (Fig. 6A). It is of interest, however, that a significant minority of Mel-10, Swiss 3T3 and QGY-7703 cells have anomalously high protein content. The whole-cell protein:NA ratio for the different cell types is plotted in Figure 6B. The different cell types separate according to absolute protein mass much as would be expected based on their 2D image area (Table 1). However there is larger variability in NA content after normalization to image area. For example, QGY-7703 cells have half the total NA content as Mel-10 and Swiss 3T3 cells, despite having nearly the same 2D area and protein mass. Somewhat surprisingly, the boxplots show that a simple, protein:NA ratio is sufficiently distinct to differentiate between cell lines. This suggests a possibility of phenotyping based on this ratio and hints at a potential role for deep-UV mass mapping in histology and diagnostics. The data also underscore the obvious but often ignored fact that, in wide-field microscopy, cell area is not a good proxy for cell mass.
Figure 6. Nucleic acid and protein scale proportionally, although protein:NA ratios are specific for cell type. A: Scatter plot showing paired nucleic acid and protein mass across six cell types: Jurkat E6(orange), CHOK1 (green), S3T3 (purple), QGY-7703 (brown), blastomere (light blue) and Mel-10 (red). Regression line (red) was calculated across all plotted cells (n = 1896 cells). B: Protein to NA ratios were calculated for individual cells of each cell type, with their distributions summarized into boxplots. The edges of the box denotes 25% and 75% percentile bounds for the sample; dashed line denotes minimum and maximum range of values not considered “outliers”. The number of cells is listed below each cell type.
Download figure to PowerPoint
Nucleic acid and protein variability in mouse blastomeres.
We analyzed eight-cell mouse embryos with the idea of testing the homogeneity of an ideally synchronized system that had progressed only three divisions after fertilization. It is fairly routine to remove one of the eight blastomeres for testing at this stage during in vitro fertilization with no serious consequences to the fetus. Therefore these cells should be a perfect model of functional equivalence.
Figure 7 presents mass maps for the blastocysts taken from seven embryos from the same dam after removal of the zona pellucidas. The embryonic cells are much larger than the somatic cells, with total protein masses of 1–2 ng and have a much higher protein to NA ratio (∼10X). There is a fair amount of well-defined structure in the nucleus, including nucleolar structure in both the NA and protein. This includes several ring structures around the nucleolar regions evident in both the NA and protein mass maps.
Figure 7. Detail of eight cell mouse embryos. A total of 54 individual blastomeres were imaged, from 8 embryos, zona pellucida removed. A representative field of view is shown for nucleic acid (A) and protein (B) mass maps. C: Scatter plot of whole cell nucleic acid and protein masses for individual blastomeres. Whole cell ROIs were manually generated. Cells from each of the eight embryos are coded with a specific color. D: Paired nucleus-isolated, NA, and protein masses were measured and plotted, using automated segmentation from matching Hoechst 33342 images. E: Paired nucleolus-isolated NA and protein masses are plotted. Nucleoli were determined by visual inspection as the low DNA regions within nuclei defined by Hoechst staining.
Download figure to PowerPoint
When pooled across seven embryos, the masses of individual blastomeres showed a standard deviation of 22% of the mean. For small, relatively homogeneous cells such as cRBCs (above), we have confirmed an approximate upper bound on the technical noise of our system (i.e. CV = 0.15). However, for the much larger mouse blastomeres (presented as nearly flat samples trapped under cover glasses), the expected errors are much lower—in our estimation on the order of 5%. We believe the data of Figure 7 reflect the true variance of the distribution with only a small component of technical noise (Table 2 for aggregate data summary). Blastomeres from an individual embryo (coded the same color in Fig. 7) have whole-cell NA and protein values dispersed over nearly a factor of two. The nuclear compartments of the eight blastomeres in each embryo have a tighter distribution, but NA and protein values also span a 1.5-fold variation in the full nucleus. Finally, the nucleolus exhibits a twofold variation in mass within blastomeres from the same embryo. There is a higher protein content in the nucleolus, approaching the same ratio as is in the cytoplasm (∼10 times more protein (peak: 17.2 pg; mean ± s.d.: 19.9 ± 6.4 pg) than NA (1.7 pg; 1.8 ± 0.4 pg) mass, within the nucleolus).
It is of considerable interest that the cells of each embryo are grouped by NA:protein ratio. This can be proven formally by forcing a regression fit through the origin in the plot of NA versus protein for the cells grouped by embryo. But the trend is seen by eye from the apparent linear distribution that separates each embryo (data of a particular color) from the aggregate data presented in Figures 7C–7E. NA:protein ratio seems to be an identifying characteristic that separates each embryo from the litter average even at this very early stage of development. It would be interesting to explore the epigenetic and genetic origins of this early differentiation.