Multidomain morphogen processing in limb bud signaling centers implements inverse morphogenesis

Significance Computer-aided morphogenesis builds explicit forward models of how genetic and biophysical programs sculpt three–dimensional anatomy. The present study suggests that higher-order patterning hubs in the vertebrate limb bud may have internalized an algorithm that runs the generative pipeline in reverse, mapping distributed molecular cues to a unified representation of the target limb shape. Thus, explicitly inferring the intended 3D structure—implemented via inverse morphogenesis—may constitute a distinct computational-level goal of embryonic patterning. This stands in sharp contrast to currently dominant approaches that emphasize the extraction of surface-level gradient statistics and which, as we show, offer poorer explanations of real tissue dynamics. Abstract Stimulus-driven, multidomain signaling in the developing limb bud is critical for converting local molecular information into coherent anatomical form. What are the representational formats of these intermediate states and how are they computed across the nodes of the patterning network? A growing literature in developmental biology focuses on the objective of acquiring high-level gradient statistics sufficient for categorizing cell fates. Here, inspired by classic theories of organ morphogenesis, we propose an alternative. We demonstrate that inferring the final 3D limb architecture may be a distinct computational-level objective, implemented via an algorithm analogous to generative simulations of how gene regulatory and biomechanical processes build tissue, but executed in reverse order. Using formation of the appendicular skeleton as a case study, we show that inverse morphogenesis spontaneously emerges in inference networks trained to map spatial signaling inputs to limb geometry. Remarkably, this correspondence to the reverse of a forward morphogenetic model also holds across the endogenous signaling domains of the vertebrate limb bud. Finally, such inference networks recapitulate the feedforward progression across the apical ectodermal ridge, zone of polarizing activity, and distal mesenchyme, outperforming current statistical-gradient models—both supervised and unsupervised—which do not align with the reverse generative logic. These findings nominate inverse morphogenesis as a multidomain algorithm embedded within embryonic patterning modules and suggest new avenues for replicating vertebrate regenerative capacity in engineered tissues.