Single-particle cryo-EM has recently emerged as a tool for characterization of protein structure at atomic resolution, with applications both to soluble and membrane proteins. Cryo-EM relies on computational alignment of single-particle images, permitting calculation of interpretable two-dimensional averages and three-dimensional reconstructions. Prior to data collection, cryo-EM samples are trapped in vitreous ice by plunge freezing, protecting biological macromolecules from radiation damage and high-vacuum, yet maintaining hydration and native structure. Soluble proteins are thus surrounded by water, which has a small electron scattering cross-section relative to the protein itself. However, membrane proteins in detergent or lipid nanoparticle systems (nanodiscs, saposin), are embedded in contrastive and predominantly unstructured material. The impact of this contrastive, unstructured mass on image alignment and three-dimensional reconstruction is unknown. Indeed, traditional cryoprotectants (glycerol) are not used in cryo-EM because their protein-like scattering is deleterious to particle contrast. Nevertheless, a growing number of membrane proteins have been characterized using cryo-EM. To date all such proteins bear large, soluble domains (often dwarfing the membrane bound component, e.g. nanodisc-anchored ribosomes), which are hypothesized to drive image alignment. Recent structures of TRPV1 (a tetrameric Ca2+ channel) reconstituted in lipid nanodiscs show that current cryo-EM pipelines can deliver high resolutions even when the ratio of structured to unstructured mass falls as low as 2:1. Here, we undertake to ascertain the limits to this ratio (if any), as well as the feasibility of reconstructing structures lacking any soluble extension.Starting from micrographs of TRPV1 in lipid nanodiscs, we use signal subtraction to simulate data for progressively smaller ratios of structured to unstructured mass, eventually eliminating all contributions from the soluble domains. Our results, including a 3.4 Angstrom structure of a hypothetical membrane protein with no soluble domain, suggest that high-resolution alignment and reconstruction is feasible for membrane proteins with no soluble extension, including when the ratio of structured to unstructured mass is less than 1:1. With continued technological development, it may be possible to adopt cryoprotectants that greatly enhance the formation of vitreous ice even during slow freezing. Finally, this work also serves as a tutorial for precise signal subtraction using open-source software, including creation of a suitable subtraction map, simulation of particle image contributions, recentering of subtracted particles, and per-particle normalization based on Fourier correlations.