Immune niches in brain metastases contain TCF1+ stem-like T cells, are associated with disease control and are modulated by preoperative SRS

Abstract The CD8 + T-cell response is prognostic for survival outcomes in several tumor types. However, whether this extends to tumors in the brain, an organ with barriers to T cell entry, remains unclear. Here, we analyzed immune infiltration in 67 brain metastasis (BrM) and found high frequencies of PD1 + TCF1 + stem-like CD8 + T-cells and TCF1- effector-like cells. Importantly, the stem-like cells aggregate with antigen presenting cells in immune niches, and niches were prognostic for local disease control. Standard of care for BrM is resection followed by stereotactic radiosurgery (SRS), so to determine SRSâ€™s impact on the BrM immune response, we examined 76 BrM treated with pre-operative SRS (pSRS). pSRS acutely reduced CD8 + T cells at 3 days. However, CD8 + T cells rebounded by day 6, driven by increased frequency of effector-like cells. This suggests that the immune response in BrM can be regenerated rapidly, likely by the local TCF1 + stem-like population.

. Sections were counterstained with DAPI according to manufacturer instructions (Thermo-126 Fisher). (2) Immunofluorescence antibody staining was performed using the Opal 7-color 127 immunofluorescence kit (Akoya Biosciences) endogenous peroxidase activity was quenched by 128 microwave treatment of the slides with AR buffer. Non-specific binding was blocked with 129 blocking/Ab diluent. After incubation with the primary antibody, the slides were incubated with 130 HRP Ms+Rb secondary antibody and then incubated in the appropriate opal fluorophore for 10 131 minutes, until staining developed. The slides were finally counterstained with DAPI. 132 133 Image capture and analysis: The selected fluorophore panel (1) allowed for simultaneous 134 visualization of three targets and a nuclear stain (DAPI) using a Zeiss Axio Scan.Z1 Slide Scanner 135 equipped with a Colibri 7 Flexible Light Source. Zeiss ZenBlue software was used for post-136 acquisition image processing. Slides stained with the Opal IHC Kit (2) were scanned using a Perkin 137 Elmer Vectra Polaris and allowed for simultaneous visualization of six targets and a nuclear stain. 138 For brightfield imaging, slides were scanned using a Hamamatsu's Nanozoomer slide scanner.
Images were analysed using CellProfiler, QuPath, and custom R and python scripts, as previously 140 described. 7 This analysis pipeline allowed for determination of the x,y location of each cell in each 141 image, as well as the quantitation of the distance between each cell type and the density of each 142 cell type. Immune niches were defined as 100um x 100um cellular neighbourhoods where both 143  Table 5. Live/dead staining was done using fixable near-IR or aqua dead cell 159 staining kit (Invitrogen). Cells were permed using the FOXP3 Fixation/Permeabilization kit 160 (eBioscience) for 45 minutes with fixation/permeabilization buffer at 4C and stained with 161 intracellular antibodies in permeabilization buffer for 30mins at 4C. Samples were acquired on a 162 Symphony instrument and analyzed using FlowJo (v10). 163 164 scRNA seq: Single cell suspensions were stained and sorted on the Beckton Dickinson FACS Aria 165 II Cell Sorter on CD45+ CD8and CD45+ CD8 + . These two sorts were then mixed 1:1 with goal 166 of enriching for the infiltrating CD8 T cell population. Single cell RNAseq libraries were made 167 using the Chromium single cell 5' Library and Gel Bead Kit (10x Genomics) and captured into the 168 Gel Beads-in-emulsion (GEMs). After the reverse transcription GEMs were disrupted and cDNA 169 was isolated and pooled. The barcoded cDNA was fragmented, end repair and A-tailing was done, followed by sample index PCR. The purified libraries were sequenced to 50,000 reads/cell on a 171 HisSeq300 (Illumina) with 26 cycles for read 1, 8 cycles for index (i7) and 91 cycles for read2. 172 173 Cellranger v3.1 was used to align, filter, count the barcodes and unique molecular identifiers 174 (UMI). Data was then analyzed using Seurat v3.0. Briefly, cells with less than 5% mitochondrial 175 genes were used. Cells that expressed less than 200 genes or more than 2000 were excluded from 176 analysis. Raw counts were then normalized for each UMI based on total expression, scaled by 177 multiplying by 10,000 and then log transformed. Variable genes were determined based on average 178 expression and dispersion, then used to perform principal component (PCA) analysis. Selected 179 PCAs were used to generate clusters and UMAP plots. Heatmaps were generated using scaled 180 expression data of marker genes, using the FindAllMarkers function in Seurat. Normalized gene 181 expression data was also shown as feature plots. Gene set scoring was performed using VISION 182 R package V2.1. Proliferation index was done as previously described. 34  To first characterize T cells in BrM, we performed flow cytometry on freshly resected tissue from 199 7 patients ( Figure 1A). To identify antigen reactive cells, we gated on PD1 + CD45RAcells. 200 Around two thirds of the CD8 + cells expressed these markers, confirming their antigen reactivity 201 ( Figure 1B). Within this PD1 + subset, we could identify both a TCF1 + Tim3stem-like cell, and a TCF1 -Tim3 + cell ( Figure 1B). Consistent with our observations in other human tumors outside 203 the CNS 7,8,12 , the stem-like T cells expressed higher levels of CD28, TCF1 and CD127 and lower 204 levels of Tim3 and CD39 compared to terminally differentiated CD8 + T cells ( Figure 1C). 205

206
To analyze a more comprehensive cohort of patients, we collected 67 tissue samples from brain 207 metastases resected at Winship Cancer Institute of Emory University (Figure 1A, Supplemental 208 Table 1). These patients had non-small cell lung cancer (NSCLC), melanoma, or breast cancer 209 and none had received prior immunotherapy (Supplemental Table 1). From these samples, we to define the overall immune architecture of the BrM (Figure 1D, E). We found a high degree of 212 variation between patients in the infiltration of CD8 + and CD4 + T cells as well as MHCII + antigen 213 presenting cells (APC) ( Figure 1F). As with the cells analyzed by flow-cytometry, the majority of 214 both CD8 + and CD4 + T cells were PD1 + indicating these cells were actively responding to antigen 215 ( Figure 1E). In these BrM, approximately 40% of the total CD8 + T cell population were TCF1 + 216 (Supplemental Figure 1D). Of note, the frequency of CD8 + TCF1 + stem-like T cells (of DAPI + 217 cells) correlated with the frequency of total CD8 + T cells (R 2 =.6233, p<0.001) demonstrating that 218 with an increase in stem-like CD8 + T cells there is a concomitant overall increase in BrM CD8 + T 219 cell infiltration ( Figure 1G). 220

221
In prior work, we had found that TCF1 + CD8 + T cells were not randomly distributed in tumors, 222 but instead are only found in close proximity with densely clustered APC, which we termed 223 immune niches. 7 Here, we found a correlation between the BrM density of stem-like T cells and 224 MHC-II + APCs (Supplemental Figure 1F) suggesting a similar inter-relationship. Next, using 225 our digitized reconstruction of whole slide IF images, we generated contour maps of the density 226 of MHC-II + cells with green dots showing the location of TCF1 + CD8 + T cells, and red dots 227 showing the location of TCF1 -T cells ( Figure 1H). Stem-like T cells clearly resided in areas of 228 higher MHC-II + cell density compared to the TCF1cells ( Figure 1H). Quantification of the 229 distance of each TCF1 + CD8 + T cell to the nearest neighboring MHC-II + cell found stem-like CD8 230 T cells were far closer on average to APCs ( Figure 1I). The TCF1 -T cells, in contrast, were 231 distributed throughout the tumor without any preference for the APC zones. These findings 232 confirm the formation of an immune niche in BrM, similar to non-CNS sites ( Figure 1J). Notably, prior data suggests that the niche is important for supporting and maintaining the CD8 + T cell 234 response within non-CNS tumors. 12 Similarly, here we demonstrate a correlation between immune 235 niche density and the frequency of total CD8 + T cell infiltration, again suggesting that these niches 236 have a supportive role in BrM ( Figure 1K). 6 Overall, these data indicate the metastasis to the CNS 237 are infiltrated by distinct functional subsets of CD8 + T cells: a TCF1 + stem-like population, and a 238 TCF1effector-like population. Furthermore, these data indicate that although these metastases 239 have entered an immunological environment where T cell access can be limited, a T cell response 240 is successfully mounted that is consistent with other solid tumor types. Local BrM recurrences are a complex problem often with limited treatment options. 35 Therefore, 245 a prognostic biomarker which could identify specific patients at highest risk for local recurrence 246 would allow for early treatment escalation. In other tumor types outside the CNS, we previously 247 showed that both a high density of total CD8 + T cells and the immune niche were associated with 248 longer progression free survival. 7 We, therefore, investigated whether these were similarly 249 prognostic for local control in BrM. First, we assessed the degree of CD8 + T cell variability across 250 all 67 samples and found substantial differences between BrM ( Figure 1F). Next, we evaluated 251 whether high CD8 + T cell density was associated with longer local control using a competing risk 252 analysis. This approach allowed us to account for the competing events of death and local 253 recurrence. In contrast to tumors outside the CNS, high CD8 + T cell density was not associated 254 with longer local control (Supplemental Figure 2A). This suggests this one general cell type may 255 not fully capture the complex interplay of different immune cells involved in BrM control. 256 257 Next, we turned our attention to the immune niche and also found significant variability in the 258 niche density across BrMs (Supplemental Figure 2B). In contrast to bulk CD8 + T cells, BrM with 259 a higher frequency of stem-like T cells or with a higher density of immune niches had longer local 260 control (Figure 2A, B). In a representative highly infiltrated BrM, there were many areas of high 261 T cell density and high APC density, and importantly, many areas where stem-like T cells and 262 APCs co-localized, forming immune niche clusters that were distributed throughout the BrM 263 ( Figure 2C, D). Following treatment, the resection cavity remained free of local recurrence at 10 264 years ( Figure 2E). In contrast, a representative poorly infiltrated BrM ( Figure 2F) showed evidence of local recurrence 6 months from the end of treatment, lacked a significant density of 266 immune infiltration, and importantly, lacked the widespread presence of these intratumoral 267 immune niches ( Figure 2G, H). The immune niche, therefore, has potential as a prognostic 268 biomarker as it captures a link between the immune microenvironment and patient outcomes which Importantly, our group has been a pioneer in the pSRS and has described several potential clinical 279 benefits of such a practice. 37 The ability to optimally sequence these treatment modalities is notable 280 given the known immune-stimulatory activity of radiation. 38 Currently, it is unknown whether 281 pSRS has similar immunostimulatory activity in the brain and what impact it may have on the 282 immune niche. 283 A total of 76 patients who received pSRS were analyzed by quantitative IF and 7 patients who 284 received pSRS were analyzed by flow cytometry. These BrM were primarily lung and melanoma. 285 The median time from pSRS to surgery was 2 days with a median dose of 15 Gy (Supplemental 286 Table 3). The pSRS and SOC cohort patient characteristics were, overall, very similar 287 (Supplemental Table 4). 288 In the pSRS BrM, we again identified PD1 + stem-like and terminally differentiated CD8 + T cells 289 by flow cytometry (Figure 3B, C). The terminally differentiated and stem-like T cells following 290 pSRS appeared phenotypically similar to SOC, with terminally differentiated cells continuing to 291 express robust levels of Tim3, CD39 and lower levels of TCF1 and CD28 (Supplemental Figure  292 3A, B). Quantitative flow cytometry analysis of pSRS BrM, demonstrated a strong trend towards 293 a decrease in the overall CD8 + T cell frequency with no change in the frequency of PD1 + CD8 + T 294 cells ( Figure 3D, E). Additionally, no significant changes were observed in the frequencies of 295 stem-like or terminally differentiated subsets ( Figure 3F, G). 296 In our larger cohort analyzed by quantitative IF, similar to SOC BrM, we could identify TCF1 + 297 stem-like CD8 T cells in the tumor tissue (Figure 3 H, I) and identified a strong correlation 298 between CD8 + T cell and stem-like CD8 + T cell frequencies (Supplemental Figure 3C). 299 Consistent with our flow data, a lower density of CD8 + T cells was observed in the pSRS vs SOC 300 cohort ( Figure 3J). Evaluating the individual subsets, we found a small, but significant difference 301 in density of the TCF1 + T cells population as well as a larger magnitude difference in the TCF1 -302 cell population (Figure 3K, L). There was, however, no difference in the MHC-II + density 303 between SOC and pSRS ( Figure 3M). Finally, while niche density was slightly attenuated 304 following pSRS compared to SOC BrM controls ( Figure 3N), the correlation between niche 305 proportion and CD8 + T cell density was maintained ( Figure 3O). These data suggest that while 306 pSRS may have differential effects on T cell subsets, MHC-II + myeloid cells may be less affected 307 and, importantly, intratumoral immune niche organization is maintained. 308

309
The transitory CD8 + T cell population is preferentially reduced following pSRS 310 311 To more specifically determine the impact of pSRS on T cell subsets, we evaluated the CD8 + and 312 This stem-like cluster also expressed low levels of effector molecules including GZMB and PRF1. 320 Cluster 2 expressed high levels of FOS, JUN and lower levels of both effector molecules GZMB, 321 PRF1 and stem-like molecules TCF7, CCR7 consistent with a transitory phenotype between 322 Cluster 1 and 3, analogous to those seen in other tumor types. 8,39 Cluster 3 (the terminally 323 differentiated effector-like cluster) was characterized by high PRF1, GZMB, and HAVCR3 324 expression and low expression of the stem-like markers ( Figure 4B). Clonotypic analysis confirmed significant TCR overlap between these 3 clusters further supporting a previously 326 described lineage relationship ( Figure 4E). 8 In the pSRS BrMs, all 3-antigen experienced CD8 + T 327 cell clusters were also identified ( Figure 4A, C). Notably, the transitory cluster was found at a 328 much lower frequency in the pSRS than in the SOC BrMs ( Figure 4C, D). The terminally 329 differentiated subset, in contrast, demonstrated increased frequency in the pSRS compared to SOC 330 BrMs ( Figure 4D). The immunodominant TCR clone was also enriched in the terminally 331 differentiated cluster in the pSRS compared to SOC ( Figure 4E). 332 333 Cells actively dividing are known to be the most radiosensitive 40 , and preclinical results have 334 shown the transitory cells to be the most proliferative of the 3 subsets. 39 Here, we also found that 335 the transitory subset had a higher proliferation score than either the stem-like T cells or terminally 336 differentiated effectors which likely accounts for their preferentially reduced frequency in the 337 pSRS cohort ( Figure 4F). We also compared the stem-like, transitory, and terminally 338 differentiated populations between the SOC and pSRS and found notable transcriptional changes 339 (Supplemental Figure 4A). In the stem-like population following pSRS, TXNIP was significantly 340 upregulated. TXNIP is associated with the response to reactive oxygen species, suggesting this 341 may be a mechanism for stem-like T cell survival after exposure to the pSRS insult. 41 Further 342 investigation is needed. 343 344 Next, we examined the other infiltrating immune cells and found a trend towards an increase in 345 the frequency of both DC and macrophages following pSRS (Supplemental Figure 4B). This 346 finding is likely due to their relative radioresistance and ability to withstand the cellular stress 347 imparted by exposure to radiation therapy. 42 Notably, we found that pSRS promotes a Type I 348 interferon (IFN) response phenotype in these two APC types (DCs and macrophages), with 349 significant upregulation of both interferon regulatory factor 5 (IRF5) and IRF8 (Supplemental 350 Figure 4C). The Type 1 IFN response is associated with maturation of DCs, and appropriately 351 provided co-stimulation is known to be critical for stem-like T cell activation and acquisition of 352 effector function. 43 In summary, pSRS has a broad impact on a diverse array of different cells. Its 353 specific effect on CD8 + T cells is subtype dependent with the most potent depletion seen in the 354 transitory population, followed by a more modest impact on the stem-like subset while the 355 terminally differentiated effector-like cells were the least affected. 43,44 356

Temporal changes of BrM immune niche components following pSRS 358 359
To further investigate the relative persistence of the terminally differentiated effector-like cells 360 following pSRS, we performed a kinetic analysis evaluating the changes in pSRS treated BrM. In 361 patients who had BrM resected on the same day as the pSRS, there was no significant difference 362 in the number of total CD8 + , TCF1 + stem-like or TCF1cells in the tumor compared to SOC 363 (Supplemental Figure 5A-C). In comparison, tissue resected between 1 and 5 days post pSRS 364 had significantly lower numbers of total, TCF1 + stem-like, and TCF1terminally differentiated 365 CD8 + T cells (Figure 5A-C). However, after 6 or more days, while the TCF1 + stem-like population 366 remained depressed, the TCFterminally differentiated cells had returned to baseline levels. Over Together with the scRNAseq data in Figure 4, these data suggest that there is a broad depletion of 374 T cells following radiation. The relatively quiescent TCF1 + stem-like cells have not yet recovered 375 even by day 6 but given the extensive data in pre-clinical models suggesting these intra-tumoral 376 stem-like cells are the source of effector-like cells in the tumor, we conclude that the recovery in 377 the terminally differentiated cells seen at D6+ following pSRS is likely driven by these remaining 378 stem-like cells. Overall, these data indicate that while pSRS does have an impact on intra-tumoral 379 T cell populations, they remain functionally able to generate an anti-tumor immune response. In this study-the largest such study to our knowledge evaluating BrM immune architecture-we 384 sought to understand the tumor immune microenvironment of BrM following up-front tumor 385 resection (SOC) or pre-operative SRS. We found that immune niches (similar to those previously 386 described in primary renal cell carcinoma, consisting of antigen presenting cells and stem-like T 387 cells) were present at varying densities in BrM under both treatment conditions (Figure 1, 3). 7 388 Remarkably, similar to tumors outside the CNS, high BrM niche density was associated with 389 longer local control of disease (Figure 2). In the pSRS cohort, there was a reduction in total BrM 390 CD8 + T cells and both the TCF1 + and TCF1subsets relative to SOC (Figure 3). Notably, the 391 immune niche organization was maintained following pSRS (Figure 3). By scRNA seq, pSRS 392 most potently depleted the proliferating transitory population while the terminal effectors were 393 less affected (Figure 4). Finally, when the pSRS BrMs were evaluated by time of resection after 394 pSRS, the total CD8 + T cell population demonstrated a rebound by day 6+ driven by an increase 395 in the frequency of the TCF -CD8 + T cells (Figure 5). The pre-operative SRS findings reported here were highly novel and intriguing to us. We had 413 hypothesized, based on our preclinical studies, pre-operative SRS would increase the density of 414 stem-like T cells and the presence of immune niches around day 7 following radiation. 36 However, 415 in our kinetics analysis in Figure 5, we did not find an increase above baseline by day 6+, but we 416 did find a rebound in total CD8 + T cell numbers that appeared to be driven by an increase in the 417 frequency of the terminally differentiated subset. These results suggest that SRS, like anti-PD1 therapy, may drive both the proliferation and differentiation of stem-like T cells into terminally 419 differentiated, effector-like cells. 5,13 This mechanism would account for the rebound in total CD8 + 420 T cells and the relative frequency changes in stem-like and terminally differentiated, effector-like 421 cells. Importantly, the newly generated effectors following pSRS may be superior to the baseline 422 TCFeffectors at controlling disease suggesting that the rebound at day 6+ may confer a clinical 423 benefit beyond a return to baseline CD8 + T cell numbers. 45 Further in vivo and in vitro functional 424 studies are necessary to confirm this hypothesis. 425

426
The larger significance of this clinically important study should not be understated. Broadly, these 427 findings demonstrate that despite the brain having several barriers to T cell infiltration, an immune 428 response is capable of being mounted that resembles that of many other tumor types/locations and 429 benefits patient outcomes. Clinically, these data suggest that the immune niche may not only be 430 an important prognostic factor for outcomes, but also be predictive of an intracranial response to 431 immunotherapy given the known importance of both stem-like CD8 + T cells and co-stimulation 432 for a robust response to anti-PD1 therapy. 5,13,43 These results also provide novel insight into the 433 optimal timing of surgical resection following SRS and integrating SRS with immunotherapy. Our 434 data indicates that resection of BrM <6+ days following pre-operative SRS may limit the 435 immunostimulatory benefit of SRS and potentially reduce the local control benefits due to the 436 acute decline in CD8 + T cells. Additionally, administering anti-PD1 therapy at the T cell nadir 437 likely also blunts the potential synergy of combinatory SRS and anti-PD1 therapy. These data can, 438 therefore, be used immediately to help guide intracranial BrM management and inform future 439 clinical investigation into optimal sequencing and combination of multiple therapeutic modalities. 440 Future clinical studies are needed to further elucidate the immune-stimulatory potential of SRS 441 and whether an intra-cranial abscopal response can be generated.      Validation of quantification methods, demonstrating correlation between a given cell count per mm 2 and respective proportion of total cells. F) Correlation between percent CD8 + of total cells and percent TCF1 + of CD8 + T cells, correlation between CD8 + cells per mm 2 and MHC-II + cells per mm 2 , correlation between CD8 cells per mm 2 and TCF1 + CD8 + cells per mm 2 , correlation between MHC-II + cells per mm 2 and TCF1 + CD8 + cells per mm 2 . G) Flow cytometry gating strategy.