Transformation-induced changes in the DNA-nuclear matrix interface, revealed by high-throughput analysis of DNA halos

In higher eukaryotic nuclei, DNA is periodically anchored to an extraction-resistant protein structure, via matrix attachment regions. We describe a refined and accessible method to non-subjectively, rapidly and reproducibly measure both size and stability of the intervening chromatin loops, and use it to demonstrate that malignant transformation compromises the DNA-nuclear matrix interface.

Establishing settings for classification of Ia and Ib by HIM. Histogram shows the percentage of the captured RN (local maxima less 75) that is also captured when using the indicated absolute intensities for RN, for a representative set of MSC1 images. f) Histogram shows classification of MSC1 images using the indicated HIM parameters compared to visual assessment of the training set. Images that were visually scored to be class Ia but by HIM were scored to be class Ib are defined as 'false Ib', and those visually scored to be Ib but by HIM scored as Ia are defined as 'false Ia'. RN 220 most accurately separates the classes. g) As e) for representative MSC1, MSC3 and MSC5 images, using HIM RNx75/RN220. h) Histogram shows mean RN radius (for class Ia and b) across the MSC series using HIMx75/15, also used to generate the data in Fig. 2c.      I  II   III  III   II  I   I  I I I I I I I I I I  I   Pixel intensity   0   50   100   150   200   250   300   0  8100  16200  24300  32400  40500  48600  56700  64800  72900  81000  Pixel intensity   I  II   III  III   II  I   RN   Distance across halo   I  II  III III  II     RN edge determination In ImageJ, thresholds can be set based on absolute intensity values or at an intensity relative to the local maxima for each individual image. Due to the wide range of maximal intensities within a population of MFHR processed cells, an absolute pixel intensity threshold is usually not applicable to the whole population. In fact, the intensity range within an RN is more similar in a halo population than is the maximal ( Supplementary   Fig. 1a), meaning an RN threshold related to the individual maxima is useful. To set the RN threshold we compared RN radius measurements returned by HIM with different settings, to visual assessment of the RN boundary (Fig. 1c, Supplementary Fig. 1b), for training images.
Choice of measurements MFHR image files were analysed in ImageJ using HIM, as described in the User Guide. HIM allows separate radius measurements to be derived from the RN and the total area ( Supplementary Fig. 1f), and therefore the calculation of a derived halo radius. Radius measurements are calculated by fitting HIM output areas to circles (User Guide). However, we note that a significant minority of entities are elliptical.
Application of an elliptical formula does not significantly change halo radius measurements (not shown), so we chose to perform circle formula based calculations throughout. Circles allow the whole area to be taken into consideration rather than just x-and y-axes, which like single visual radius measurements, are more affected by structural irregularities. In most cases we choose to present our results in terms of derived halo radius, as this reflects the theoretical chromatin loop size and has biological implications. If desired, and perhaps applicable to certain specific analyses, HIM allows different types of measurements to be made quickly and easily. For example RN radius or area measurements could be used to interrogate NM volume and compaction. Note that our standard combination of settings (HIM x55/15) systematically incorporates pixels at the outer halo edge that are not visible by eye ( Fig. 1e), meaning halo measurements are systematically larger than those generated by eye.
Effect of flare The effect of flare from a bright RN on halo measurement was assessed using unexpanded 3T3 cell halos (treated only up to 0.5 M NaCl), in which histones are not extracted 1 , and chromatin remains packaged within the nucleus. HIM x55/15 returned a halo radius that is approximately 5-fold less than that created by loop expansion, indicating a small but significant, systematic contribution to halo measurements (Supplementary Fig.   1g).
Using HIM to determine classifications Specialist HIMs were designed which use RN intensity to non-subjectively classify halos. Populations of cells processed by MFHR return two classes of product (Fig. 1h). Class I have defined RNs and class II have an ill-defined RN with poor structure and fail to return a value using HIM. Class I is further divided into class Ia (bright RN) and Ib (pale RN, Fig. 1h). Class Ia halos are defined as those in which >50% of the returned RN from local maxima is above 220 intensity. Class Ib halos have <50% of their RN above 220 intensity. Validation is shown in Supplementary Fig. 2a. RN HIM was trialled with different absolute intensity values (180-250 Supplementary Fig. 2a) and 220 chosen. For an example training set of visually defined Ia and Ib images, HIM RNx55/RN220 returned a value of 89% for class Ia, but only 7% for class Ib. Therefore using a cut off of 50% HIM RNx55/RN220 effectively separates class Ia and Ib halos and we show that HIMRNx55/RN220 classifies the whole 3T3 dataset the same as visual scoring ( Supplementary Fig. 2a right). Nevertheless, other cell lines are less easy to classify by eye so RN HIM provides rapid, non-subjective classification by application of a constant intensity threshold.  Fig. 2b-d), implying global differences in the frequency or nature of their MARs. We considered whether changes in halo size might reflect differing proportions of classes across populations, as we find that class Ia cells return slightly larger RN radius values and slightly smaller halo radius values than Ib cells using standard HIM x55/15 (Supplementary Fig. 2b). However the differences we observe in halo size across cell lines is not attributable to differing distribution within class I, because separate analysis returns similar trends (Supplementary Fig. 2c). In fact there is not a straightforward correlation between mean halo size and classification (compare Supplementary Fig. 2c and 2d). For example, NHU and MCF10A have a similar percentage of class Ia cells (78% and 88% respectively), but different mean halo radius (16 µm and 13 µm respectively), while BJ-hTERT has only 42% class Ia cells, and a mean halo size similar to MCF10A. The differences in halo parameters that we observe across a panel of non-cancer cell lines therefore shows that choice of 'normal' comparison is highly important.

Customization of HIM for MSCs
Cell types respond to MFHR differently, in some cases requiring HIM thresholds to be fine-tuned for particular analyses. Mesenchymal Stem Cells (MSCs) 2 were observed to produce MFHR images with pale RNs, requiring the RN threshold criteria to be adapted so that it supports analysis across the whole diverse series. Different RN values, based on range from local maxima (where x=55, 70, 75, or 100) were compared for measurement of MSC1 and MSC3 cell lines (Supplementary Fig. 3b). The HIM default threshold (x55) gave a smaller than visual measurement for both cell lines, while RN x75 gave a similar value to visual assessment (100.7% for MSC3). MSC1 RN measurements are still somewhat smaller than visual measurements at x75, which will lead to slight overestimation of halo size. In fact several sets of HIM settings were trialled on the MSC series data (Supplementary Fig. 3c). HIM x75/x180 (Fig. 2b) was used to derive a halo measure in which both parameters are related to RN intensity. For this analysis, the range settings for outer threshold (x180) was chosen to support inclusion of a greater proportion of cells across the series. Outer range settings higher than 180 incorporate smaller percentages of the population, but still report differences in halo measurements between MCS1 and 3 ( Supplementary Fig. 3c, d). A different version of HIM (x75/15, Fig. 2c) uses an outer threshold with absolute pixel intensity of 15, which reflects the point at which intensity increases over background, where background is set at 5% AUC for a training population (Fig. 1e). Here outer threshold is unaffected by RN intensity and reports only on differences in RN and halo size. We remind readers that data collected using different thresholds should not be compared. For example, mean halo measurements in figure 2c cannot be directly compared to figure 2b, except to evaluate the effect of the different settings.

Customisation of classification HIM for MSCs
To establish suitable parameters for classification of MSCs by HIM, data collected using different absolute RN pixel intensity thresholds were compared to classifications by eye for a representative set of images for MSC1 (Supplementary Fig. 3e, f). Those that were visually scored as class Ia, but by HIM were scored as class Ib, are defined as false Ib, and those visually scored as Ib, but by HIM scored as Ia, are defined as false Ia. For MSC1, classification based on a 50% cut off, using intensity above 180 or 200 caused significant percentages of false Ia images. Likewise, above 230 produced a significant percentage of false Ib images (Supplementary Fig. 3f).
We chose 220 intensity as our parameter, as this minimized 'false' classification with the training set, and confirmed that this is suitable across the full range of the MSC series ( Supplementary Fig. 3g), before application to test populations.

Validation of biological differences
As with all analysis methods that image slides prepared in different sessions, we considered whether biological differences could be seen above the level of variability between replicate experiments. To illustrate this, frequency distributions were plotted for halo size measurements generated using HIM x75/x180 from MSC cell lines, for complete data sets (Supplementary Fig. 4a) and individual biological and technical replicates (Supplementary Fig. 4b, c, d), and variance calculated using ANOVA (Supplementary Fig. 4e). Graphs show that individual replicates from the same cell line overlay well and are more similar to each other than to data sets from a different cell line, compare MSC1 and 2 (Supplementary Fig. 4b) and MSC4 and 5 ( Supplementary   Fig. 4c). Furthermore, individual data sets return the same results as the mean of each set of replicates (Supplementary Fig. 4a, Fig. 2b). ANOVA results confirm that MSC1 halo size data set is statistically different from MSC2, and that MSC4 is statistically different from MSC5, but MSC2 is not different to MSC4. Overall, this shows that variation inherent in preparation conditions is far less than biological differences between cell lines.
In addition, we processed and imaged MSC1/MSC4 and MSC4/MSC5 as pairs, in order to ensure identical reagents and process. As expected, we find the same statistical differences as when processed independently (compare Supplementary Fig. 4f and Fig. 2b). Taken together this allows us to conclude that the differences we observe between cell lines is not due to slide to slide variability but statistically robust biological differences due to cell type.  In addition, when HIM has finished processing each file within the source folder, any images where a RN and outer pair was not found for all ROIs will be listed in a dialogue box titled, "Files requiring manual curation". Images will also be listed here for which no RN, or no outer was detected.

Matching RN and outer measurements
Images which 'fail' HIM also checks any outer measurements which 'fail' (see description of stability measurements and class II cells) in that the outer measurement occupies more than 80% of the image area. This happens typically for pale cells when outer measurements are measured using the local maxima less x method eg when unstable cells decay during stability analysis. Outer measurements that are larger than 80% of the image area are replaced by 'fail' under outer area measurement heading. This is applied to all RNs on an image for which an outer area measurement is classified as a fail. Therefore, if wishing to refer back to outer threshold images users should check the image name and outer number carefully.

Results column headings
Headings are slightly different for classification HIM output with outer replaced by absolute RN. This refers to the ROI generated using an absolute pixel intensity threshold rather than one related to the local image maxima. This results in a new picture, and selects the pixels within a tolerance of 55 from the local maximal intensity of the image. The area in white will be measured.

ImageJ: Process: Binary: Close
The individual pixels selected will then be closed as an ROI.

Technical tips for MFHR processing and image generation
These are provided to help minimize other sources of variability.
• • Minimize time between visualization of a halo and image capture • Take care not to re-image areas of the coverslip that have already been exposed to microscope light.
• For class analysis no halo should be excluded from image capture. These image sets can also be used for size analysis. If only size analysis will be performed, joining halos can be excluded at image capture, since HIM cannot measure these halos.
• Exposure time, and RN and outer thresholds should be set using a training set.