Integrated analyses of single-cell atlases reveal age, gender, and smoking status associations with cell type-specific expression of mediators of SARS-CoV-2 viral entry and highlights inflammatory programs in putative target cells

Christoph Muus; Malte D Luecken; Gokcen Eraslan; Avinash Waghray; Graham Heimberg; Lisa Sikkema; Yoshihiko Kobayashi; Eeshit Dhaval Vaishnav; Ayshwarya Subramanian; Christopher Smillie; Karthik Jagadeesh; Elizabeth Thu Duong; Evgenij Fiskin; Elena Torlai Triglia; Christophe Becavin; Meshal Ansari; Peiwen Cai; Brian Lin; Justin Buchanan; Sijia Chen; Jian Shu; Adam L Haber; Hattie Chung; Daniel T Montoro; Taylor Adams; Hananeh Aliee; Samuel J Allon; Zaneta Andrusivova; Ilias Angelidis; Orr Ashenberg; Kevin Bassler; Christophe Becavin; Inbal Benhar; Joseph Bergenstrahle; Ludvig Bergenstrahle; Liam Bolt; Emelie Braun; Linh T Bui; Mark Chaffin; Evgeny Chichelnitskiy; Joshua Chiou; Thomas M Conlon; Michael S Cuoco; Marie Deprez; David S Fischer; Astrid Gillich; Joshua Gould; Minzhe Guo; Austin J Gutierrez; Arun C Habermann; Tyler Harvey; Peng He; Xiaomeng Hou; Lijuan Hu; Alok Jaiswal; Peiyong Jiang; Theodoros Kapellos; Christin S Kuo; Ludvig Larsson; Michael A Leney-Greene; Kyungtae Lim; Monika Litvinukova; Ji Lu; Leif S Ludwig; Wendy Luo; Henrike Maatz; Elo Maddissoon; Lira Mamanova; Kasidet Manakongtreecheep; Charles-Hugo Marquette; Ian Mbano; Alexi M McAdams; Ross J Metzger; Ahmad N Nabhan; Sarah K Nyquist; Jose Ordovas-Montanes; Lolita Penland; Olivier B Poirion; Segio Poli; CanCan Qi; Daniel Reichart; Ivan Rosas; Jonas Schupp; Rahul Sinha; Rene V Sit; Kamil Slowikowski; Michal Slyper; Neal Smith; Alex Sountoulidis; Maximilian Strunz; Dawei Sun; Carlos Talavera-Lopez; Peng Tan; Jessica Tantivit; Kyle J Travaglini; Nathan R Tucker; Katherine Vernon; Marc H Wadsworth; Julia Waldman; Xiuting Wang; Wenjun Yan; Ali Onder Yildirim; William Zhao; Carly G K Ziegler; Aviv Regev

Uncertainty is challenging to model in our single-cell meta-analysis as variability exists on the levels of both donors and cells. For simplicity, we modeled the overall variance with both contributions covered implicitly by treating each cell as an independent observation. As cells from the same donor cannot be typically regarded as independent observations, this can result in inflated p-values, especially when there are few donors for a particular cell type. To counteract this limitation, we employed three approaches: (1) We used a simple noise model (Poisson) to reduce the chance of overfitting donor variability to obtain spurious associations; (2) We confirmed significant associations from the single-cell model in a pseudo-bulk analysis to ensure effect directions are consistent when modeling only donor variation (Methods, Fig. 3d , Extended Data Fig. 6, 7, 8, Supplementary Data D1) ; and, (3) We investigated whether significant associations change direction when holding out any one dataset to ensure that the effect is not dominated by the inclusion of many cells from only one source (Methods, Fig. 3f, Supplementary Data D1) . We regarded an association that passes all of these validations as a robust trend, while associations that appear dominated by a single dataset (often because this dataset is a major contributor of a given cell type) were denoted as indications.