Have you measured the background of your blot and subtracted this from your band values using a similar size box?

Michael john Fry Yes I did

I think it's as simple as you're approaching the detection limits of this readout, whatever the combination.

What chemiluminescent substrate are you using and what is the published linear range?

What background removal method did you use? Michael's suggestion implies an image rectangle style background removal style which only works if background values (typically averaged inside the defined AOI) are even across the entire image which very rarely happens in real life. This would explain why you get closer to a 0 intersect with higher exposures where your background has basically all been washed out to white (so will be very even in terms of imaged pixel values).

I guess the next question is why do you want it to pass through 0? If you're trying to determine linearity of antibody response, it's important that this is typically only necessary to perform across the dynamic detection range of the antibody rather than proving with 0 samples you get 0 response.

Michael Ronzetti and Steven Dodd, many thanks for your comments. Below I answer your questions:

Michael

We are using  Clarity Western ECL Substrate BioRad 170-5061 substrate for visualition of HRP-conjugated secondary antibody. I am not pretty sure what you mean by ‘the published linear range’ – the range for antibody of substrate? I think this range depends on concentration of ABs? ‘You're approaching the detection limits of this readout’ – do you mean that we are using too little protein and therefore we are still at the lowest part of the linear range? So we can use more protein, and if we are still within the linear range, we can assume that our line goes through origin because the relative magnitude of this error becomes negligible? (this error is relative larger if we load 4 microg of protein than when we load 30 microg, simply in the latter case we are further from zero anyway so it is not so important if we go straight through zero or we miss it a bit). Indeed, we loaded less than 10 microg of protein (our gel is 0.75 mm thick)

Steven

We typically use 4-6 ‘background rectangles’ and the calculate mean value that is subtracted from rectangles with samples. Background rectangles are distributed randomly along the line of samples and there is very little variation in their values. So we do not think the problem is here. ‘why do you want it to pass through 0? ‘ – because we liked to compare results from several gels. We use mini gels with 10 lanes, so we liked to use one lane for Kaleidoscope protein ladder, three lanes for one concentration of the same standard sample on each gel, and six lanes for analyzed samples. We liked to divide value for each analyzed sample by mean result for standard sample (three repeats to assure high reliability). But we think it is appropriate only if calibration line for several dilutions of standard (and samples) goes through origin i.e. twice higher amount of analyzed protein always results in twice stronger signal. This is not the case – and thus now we think we will load on each gel four lanes with different dilutions of standard and six lanes with analyzed samples. Thus, we will have calibration curve from the same standard done individually for each gel. It seems fine since it should give us more accurate results, the workload is the same (six analyzed samples per gel in both cases), and we do not need Kaleidoscope on each gel (we use bromophenol blue in our samples that moves faster than our UCP1 protein so we will not lose it). What do you think about this plan?