Extended Data Fig. 5. Process flow for extracting information from XRI assemblies.

Step 1: For each XRI (a), a curved centerline was drawn along the longitudinal axis of the XRI in the anti-HA channel (b). The centerline width was set to half of the width of the XRI. Step 2: The intensity profiles along this centerline were measured in the anti-HA channel (resulting in an HA line profile; cyan curve in c) and in the other XRI epitope staining channel, such as in the anti-FLAG channel (resulting in a FLAG line profile; magenta curve in c). Step 3: Next, each of the line profiles was split into two half line profiles using the geometric center point of the XRI (the 50% length point along the centerline, measuring from the end of the XRI; gray dashed vertical line in c) as the ‘split point’. Each of the half HA line profiles was then converted into a line integral of HA, by integrating the line profile with respect to the distance along the half centerline starting from the split point, and then these line integrals of HA were normalized to the maximum integral value so that each line integral of HA started at the value 0 at the split point of the XRI, and gradually increased to the value 1 at the end of the XRI (see Methods for equations for the quantifications throughout this figure). For the corresponding half FLAG line profiles, line integrals were also calculated but not normalized. At this point, we have the line integrals of HA and FLAG, which correspond to the cumulative HA and FLAG intensities along each half of the XRI. The FLAG intensity change per unit change in the cumulative HA intensity, defined as the FLAG signal, was calculated by taking the derivative of the line integral of FLAG with respect to the line integral of HA (gray curves in d). At this stage, we had obtained the line integral of HA and the FLAG signal from each of the halves of the XRI, and the final extracted FLAG signal from this XRI (black curves in d) was defined as the point-by-point average of the two FLAG signals from the two halves of the XRI. Step 4: We found the two obtained FLAG signals from the same XRI have small but noticeable differences (see the two gray curves in d). We reasoned that such small but noticeable discrepancies between the two halves of the same XRI was due to the asymmetry of the XRI, and the choice of the exact geometric center as the split point may not be optimal. To minimize the discrepancy between the two FLAG signals from the two halves of the same XRI, we searched for an optimal split point (black dashed vertical line in e) near the geometric center of the XRI (searching range was the geometric center ±25% of the total XRI length, that is, between −0.25 and 0.25 on the x-axis in e), so that using this optimal split point, instead of the geometric center, as the split point would result in the least difference (in terms of sum of squared differences) between the two FLAG signals from the two halves of the split XRI. Step 5: Same as Step 3, except that the optimal split point, instead of the geometric center, was used to split the line profiles into two halves (f). We found the resulting final FLAG signal (after averaging those from the two halves) when using the geometric center as the split point was similar to that when using the optimal split point as the split point (compare the black line in d and f). Nevertheless, we used the optimal split point as the split point to analyze XRIs throughout this paper.