SUMMARY Dopamine (DA) neurons in the midbrain ventral tegmental area (VTA) integrate complex inputs to encode multiple signals that influence motivated behaviors via diverse projections. Here we combine axon-initiated viral transduction with rabies-mediated transsynaptic tracing and Cre-based cell type-specific targeting to systematically map input–output relationships of VTA-DA neurons. We found that VTA-DA (and VTA-GABA) neurons receive excitatory, inhibitory, and modulatory input from diverse sources. VTA-DA neurons projecting to different forebrain regions exhibit specific biases in their input selection. VTA-DA neurons projecting to lateral and medial nucleus accumbens innervate largely non-overlapping striatal targets, with the latter also sending extra-striatal axon collaterals. Using electrophysiology and behavior, we validated new circuits identified in our tracing studies, including a previously unappreciated top-down reinforcing circuit from anterior cortex to lateral nucleus accumbens via VTA-DA neurons. This study highlights the utility of our viral-genetic tracing strategies to elucidate the complex neural substrates that underlie motivated behaviors.

SUMMARY Viral-genetic tracing techniques have enabled mesoscale mapping of neuronal connectivity by teasing apart inputs to defined neuronal populations in regions with heterogeneous cell types. We previously observed input biases to output-defined ventral tegmental area dopamine (VTA-DA) neurons. Here, we further dissect connectivity in the VTA by defining input-output relations of neurochemically and output-defined neuronal populations. By expanding our analysis to include input patterns to subtypes of excitatory (vGluT2-expressing) or inhibitory (GAD2-expressing) populations, we find that the output site, rather than neurochemical phenotype, correlates with whole-brain inputs of each subpopulation. Lastly, we find that biases in input maps to different VTA neurons can be generated using publicly available whole-brain output mapping datasets. Our comprehensive dataset and detailed spatial analysis suggest that connection specificity in the VTA is largely a function of the spatial location of the cells within the VTA.

Translation-based methods for grammar correction that directly map noisy, ungrammatical text to their clean counterparts are able to correct a broad range of errors; however, such techniques are bottlenecked by the need for a large parallel corpus of noisy and clean sentence pairs. In this paper, we consider synthesizing parallel data by noising a clean monolingual corpus. While most previous approaches introduce perturbations using features computed from local context windows, we instead develop error generation processes using a neural sequence transduction model trained to translate clean examples to their noisy counterparts. Given a corpus of clean examples, we propose beam search noising procedures to synthesize additional noisy examples that human evaluators were nearly unable to discriminate from nonsynthesized examples. Surprisingly, when trained on additional data synthesized using our best-performing noising scheme, our model approaches the same performance as when trained on additional nonsynthesized data.