Background/Context: One of the most critical factors for improving the quality of reading comprehension instruction provided by classroom educators is teacher professional development (PD) (Desimone, 2011). However, the COVID-19 pandemic and shift to virtual instruction significantly impacted how teachers access and engage in PD. As many recognize (Hartshorne et al., 2020), supporting sustainable change through web-based PD may be a viable option to meet the new digital landscape of professional learning. Moreover, PD delivered through an online platform may provide access to high-quality professional learning to a broader range of geographic regions that may not otherwise have such opportunities. One effective method of PD is practice-based professional development (PBPD; Ball & Cohen, 1999), which is characterized by a shift from simply delivering knowledge to focusing on the tasks and activities teachers use for instruction within the classroom (Ball & Cohen, 1999). In this study, we examine the effectiveness of a web-based PBPD that choreographed synchronous and asynchronous instruction to deepen teachers' understanding of top-level structures (i.e., comparison, cause-effect, problem-solution, sequence, description; Meyer, 1975), which is well established by the literature for supporting and improving students' reading comprehension abilities (Boegaerds-Hazenberg et al., 2020). Our focus on building teacher knowledge is grounded in the emerging evidence that improving teacher knowledge leads to improved student reading outcomes (Hudson et al., 2021). Purpose/Objective/Research Question: In this study, we addressed the following research questions: (1) To what extent are there heterogeneities in teachers' pretest and posttest understanding of top-level structure (i.e., organization, problem, cause of the problem, and solution)? (2) To what extent do teachers with a poor understanding of top-level structure at pretest transition to a group with better understanding after participation in a virtual PBPD? (3) To what extent are the asynchronous training activities used during the virtual PBPD associated with the transition from a group with a poor understanding of top-level structure at pretest to a group with a better understanding at posttest after controlling for the grade level of teaching, years of teaching experience, and certification type? Setting and Participants: Participants (N=409), recruited through a partnership with a large literacy organization from a state in the south, voluntarily attended the virtual PBPD offered during the COVID-19 pandemic (i.e., summer 2020). Participants represented over 136 schools within 59 different school districts, including larger urban districts and smaller rural districts. Participants' demographic information is presented in Table 1. Intervention/Program/Practice: All participants received two days (i.e., 12 hours) of web-based PBPD focused on the Knowledge Acquisition and Transformation (KAT; Wijekumar et al., 2012) framework, which provides direct and explicit text structure-based reading comprehension instruction for students. Over the two days, teachers participated in synchronous whole and small group virtual instruction choreographed with online, asynchronous practice activities. See Figure 1 and Appendix A for more information. Data Collection and Analysis: Two equivalent passages were given to teachers through the online platform to assess teachers' understanding of using top-level structures to generate a main idea after reading. Passages were equivalent in terms of length and readability. Each text presented a problem and its cause, as well as a solution that eliminated the cause of the problem. After reading, teachers were asked to generate the passage's main idea. Based on Meyer et al.'s (2010) approach, teachers' main idea statements were scored in each of four categories: (a) organization, (b) cause, (c) problem, and (d) solution. The four categories were measured on a 0-3 quality rating scale, with a 3 indicating the highest quality. Teachers also completed ten asynchronous online training tasks that were choreographed to support synchronous instruction. Each task was scored on a 0-100 scale. The mean score of these ten tasks was used as a training score for the subsequent analysis. Cronbach's ? for the training score was 0.96, indicating excellent reliability. For RQ1, we conducted Latent Class Analysis (LCA) at pretest and posttest, respectively, to identify potential subgroups of participants with particular patterns across their main idea statements: (a) organization, (b) cause, (c) problem, and (d) solution. For RQ2, we conducted an unconditional Latent Transition Analysis (LTA) using the posterior results from the LCA analyses to analyze the sample size and profile of each classification of teachers. For RQ3, we conducted a conditional LTA model using the mean online training score to predict posttest classification, controlling for pretest classifications and three covariates (grade level of teaching, years of teaching, and certification type). We were particularly interested in whether the transition from a group with a poor understanding to a group with a better understanding of top-level structure was associated with the asynchronous online activities used during the PBPD after controlling for other predictors. Findings/Results: The 3-class model demonstrated the best fit (see Table 2), and participants were classified into a latent class using each teachers' expected class membership (see Table 3). Classes were named by salient factors observed. As shown in Table 4, there was a significant negative association between the online training and Class 1 versus Class 3 and Class 1 versus Class 2 at posttest (logit estimation=-0.09, SE=0.03, p<0.01, -0.07, SE=0.02, p<0.001, respectively). Grade level of teaching, years of teaching, or certification type were not associated with these between-class comparisons. Conclusions: Mirroring previous studies (Beerwinkle et al., 2018; Wijekumar et al., 2020), few teachers had a good understanding of using top-level structures to support the generation of a main idea at pretest; consequently, these teachers may have faced challenges in planning for explicit text structure instruction. However, most teachers transitioned to the Good Overall Understanding class following participation in the web-based PBPD, indicating that teachers' main idea statements included key details from the top-level structure of the text. Further, teachers with lower training scores were more likely to be in the Poor Overall Understanding class than the two higher classes, suggesting that the asynchronous activities choreographed to support the synchronous web-based training played a critical role in teachers' learning. Thus, this finding suggests that web-based PBPD is an effective method for those aiming to improve teachers' understanding of evidence-based reading comprehension instruction.