5.2.1 Qualitative Observations
We imaged cells expressing GFP fusions to the representative protein set of themembrane proteome. With manual inspection of all cells, we found that all proteins were distributed non-homogeneously. Even proteins previ- ously annotated to cover the entire PM in a homogeneous manner showed network-like patterns. Interestingly, proteins of ourmembrane proteome not only localized in the previously observed distinct patch- or network- like patterns (Berchtold and Walther 2009, Malínská et al. 2003, Young et al. 2002), but also in patterns which appeared to be many intermediate variants (Figure 5.2a). For example, we observed proteins like Bio5 with distinct and equally-distributed patches, Mep2 with mostly patches with few track-like elements or patches in close proximity and Hxt3 with many track-like elements distributed adjacent to patches (Figure 5.2b). Many proteins, such as Hxt3, formed networks that had so far only been re- ported for Pma1. Notably, lipid-anchored were also not homogeneously distributed but formed equal networks (Gpa1, Ras2 and Psr1).
5.2 Quantification of Spatial Patterning 117
patch-like network-like
Fet3GFP
Mid2GFP
Sur7GFP Mep2GFP Gpa1GFP Psr1GFP Ras2GFP
Bio5 Mep2 Hxt3
Linescan
Intensity (a.u.) Intensity (a.u.)
*
Bio5GFP Hxt3GFP
Intensity (a.u.)
*
Figure 5.2: Protein Domain Patterns. TIRF microscopy of GFP-labeled PM
proteins revealed not only the two basic lateral distribution patterns – patch- like and network-like – but what appeared to be intermediate variations of the two. Representative line scans showed patch- (local maxima) or track-like (marked with asterisks) elements.
5.2.2 Image Quantification by Network Factor
To better assess the pattern diversity, we defined a network-likeness to har- bor track-like low intensity elements mostly connected to high-intensity patches in the image and assumed these elements to be generated by pro- teins below the temporal or spatial resolution.
The existence of fine-granular differences in the domain patterns called for the development of a new algorithm to quantify these images. The quantification was aimed to yield a single numeric factor for each cell’s spatial pattern capturing the characteristics. We sought to differentiate between patterns with mostly high-intensity areas (tendency to be patch- like) and patterns with a reasonable fraction of intermediate-intensity ar- eas (tendency to be network-like). The factor ideally should range between cells of “unique patch-like” up to “densely network-like”.
0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 Intensity Histogram Area Over Cummulative Histogram Curve 0.0 0.20.40.6 0.81.0 0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.20.40.6 0.81.0 0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.20.40.6 0.81.0 0.0 0.2 0.4 0.6 0.8 1.0 0.003 0.092 0.4367 Network Factor Ssy1GFP Nha1GFP Ras2GFP 0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 1.0 a 0.00 0.05 0.10 0.15 0.20 0.25 0.30 0.35
SSY1 PIL1 SUR7 BIT61 FPS1 SHO1 ATR1 MNR2 ENA1 RAX2 DNF1 ITR2 MSS4 BIO5 MEP2 ENA2 SLN1 RSN1 TCB3 FET4 YOR1 PDR12 FLC1 NHA1 FUI1 THI7
YLR413W
GAS1 VHT1 MID2 FTR1 FAA3 TPO1 HNM1 GPA1 GAP1 BAP2 FET3 PDR5 PSR1 HXT2 MRH1 HXT6 PMA1 HXT3 HXT1 PMP1 RAS2
GFP
Network Factor (a.u.)
b
Figure 5.3: Network Factor of Domain Patterns. a. From the image, first,
64-bin intensity histograms were calculated, second, cumulative histograms built and, finally, area over the cumulative histogram curves were sufficient
to derive the desired network factor. b. Comprehensive protein set was
5.2 Quantification of Spatial Patterning 119
To develop the “network factor” quantifying lateral protein domain patterns, we first generated intensity histograms. Histograms were in prin- ciple able to capture the desired characteristic: whenever cells had just patches or more and more intermediate intensity values histograms had distinguishable patterns. The number of bins was fixed to 64 while mini- mal and maximal intensity value were used as lower and upper intensity boundaries, respectively. Thus, intensitiesI= (ii j)of the ROI of a cell in x
and y were used to generate the 64 bin counts for the histogramH= (hi)
with 1 ≤ i ≤ 64 andhi the number of pixels in I of respective intensity
value falling in the range of hi. Three sample proteins were selected to
visualize the differences in their histograms for a patch-, an intermediate and a network-like pattern (Figure 5.3a).
The idea to yield one factor for one entire histogram for one cell was to first generate a cumulative histogram C = (ci)with ci = ∑ij=1hj. Al-
ready conceivable from the cumulative histograms in Figure 5.3a, a single network factor may be calculated fromCby e.g. the area over the curve. The new network factor nf was subsequently defined as
nf(I) =
64
∑ i=1
ci
When calculating the network factor for each cell for each labeled pro- tein (45 proteins withn≥10) the domain patterns visible from the initial
manual qualitative study were nicely preserved with the network factor (Figure 5.3b). The protein Sur7 and Pma1 – described to form a patch- and network-like pattern respectively – were ranging in the outer spectrum of the nf values. Interestingly, proteins, like Pmp1 (the regulatory peptide of Pma1) and Ras2 (lipid anchored Rho-GTPase) form even stronger net- works than Pma1 itself. The protein Ssy1 (amino acid sensor) showed an even more distinct patch-like pattern. Note, that the factor was named network factor while network-like values of nf→0 indicated a patch-like
pattern, thus, without track-like (intermediate intensity values) elements in the image.