Culture medium is a chemist's choice, made in a lab. Blood carries its own composition, unchosen: it arrives with the plasma proteins that stick to a lipid nanoparticle's outer fatty layer within moments of injection, and with the circulating immune cells that then decide whether to swallow the coated particle, which is what determines the effect of the medicine downstream [9]. That is the case for running the screen in blood, and it is a decent one: the variable the assay adds back is the variable cell culture takes away.
Take the 90 percent non-approval rate literally and each approved drug sits on top of roughly nine candidates that got as far as human testing and failed [8]. Some of that attrition is biology no bench assay reaches. Some of it is the species mismatch Hak points to, where mouse data flatters a formulation that a human bloodstream will ignore or reject [12]. A human-blood screen cannot touch the first category; it is aimed squarely at the second.
What the published account establishes is discrimination: the readout moves when the formulation changes [3]. What it does not supply, at least not in the account of the work, is a variant count, a donor count, effect sizes, or any comparison between the blood ranking and what the same particles did in an animal or a person [16]. Discrimination is only the necessary condition for a screen; sufficiency is a separate bar, one this account does not clear. What remains unknown is sensitivity: how often a candidate that looks bad in donor blood would have worked anyway. That number is the one a program director actually needs, because a filter with unknown false-negative behaviour quietly deletes options.
Fit between readout and product is where the method will earn or lose its keep. For a candidate whose target cells sit outside the bloodstream, immune capture is a loss channel and a tolerability signal rather than the mechanism of action. For in vivo CAR-T, the example Hak singles out [14], the intended mechanism is engagement with T cells in circulation, since the whole point is to install the cancer-recognizing receptor inside the patient rather than shipping cells out to a laboratory and back [13]. If whole blood is going to move a go/no-go call, that is where it should show up first.
Hak himself calls the test a small piece of the bigger puzzle whose value is deciding earlier whether to continue [6], and his claim for human relevance is the modest one, that findings from human blood may be more relevant than findings in experimental animals [5]. Cheap filtering of unknown sensitivity is still worth something when mRNA has to be wrapped in a particle to survive at all [10] and that particle can be assembled a thousand different ways [11], only a few of which can be put into animals. The condition on that view is one published case where the blood ranking and the in-life outcome agree.
