Mid-infrared Colors Vary with Galactic Environment: Contrasting Star-forming Disks, Young Centers, and Quiescent Star-formation Deserts

We use near- and mid-IR images from the PHANGS-JWST Cycle 1 (GO 2107, PI: Lee; J. C. Lee et al. 2023) and Cycle 2 Treasury (GO 3707, PI: Leroy; R. Chown et al. 2025, A. Leroy et al. 2026, in preparation). We focus on the diffuse ISM using five filters: NIRCam F300M (3 μm) primarily traces starlight; MIRI F770W (7.7 μm) traces C–C stretching modes of more ionized PAHs; F1000W (10 μm) traces a combination of continuum from stochastically heated small grains, silicate absorption, and PAH emission; F1130W (11.3 μm) traces C–H out-of-plane bending modes of more neutral PAHs; and F2100W (21 μm) traces the hot dust (small grains) continuum (A. K. Leroy et al. 2023; C. M. Whitcomb et al. 2023; G. P. Donnelly et al. 2024; R. Chown et al. 2025c). In Section 4.4 we also compare to maps of 3.3 μm PAH emission captured in the F335M filter for the 19 Cycle 1 galaxies. This feature traces smaller, more neutral PAHs. H. Koziol et al. (submitted) describe the production of these F335M_PAH feature maps.

The Cycle 1 PHANGS Treasury includes 19 galaxies with full coverage of all filters. We include 52 galaxies from the Cycle 2 Treasury with F300M, F770W (the shorter-wavelength PAH band), and F2100W.²⁹ NGC 4826 (Cycle 2) has additional F1000W and F1130W coverage from GO 3177 (PI Sun), which brings our full sample to 20 galaxies with full MIRI coverage, and 51 galaxies without F1000W and F1130W.

The data are reduced using pjpipe (T. G. Williams et al. 2024). Modifications to pjpipe used in the latest combined Cycle 1 and 2 data are described in R. Chown et al. (2025).³⁰ In brief, pjpipe adds modifications to the STScI JWST pipeline (H. Bushouse et al. 2025) to improve the handling of extended sources. The JWST observations are astrometrically registered against Hubble Space Telescope broadband imaging using bright point sources, which in turn have been aligned to Gaia DR3 sources.

We analyze the NIRCam and MIRI images after convolving them to a common Gaussian PSF, which is effectively beam-matched to the MUSE data resolution. For F2100W, our longest wavelength filter, this resolution also represents a compromise between improving signal-to-noise (S/N) and decreasing angular resolution (see T. G. Williams et al. 2024; R. Chown et al. 2025 for details).

2.1.1. Background Determination

2.1.2. Starlight Subtraction

The MIRI filters usually contain some stellar continuum emission in addition to PAH and/or small dust continuum emission (see, e.g., G. Helou et al. 2004; D. A. Dale et al. 2009; C. M. Whitcomb et al. 2023; J. Sutter et al. 2024). We estimate and remove contamination from the stellar continuum for each MIRI filter using the NIRCam F300M emission as a template for the stellar emission. We use the prescription from J. Sutter et al. (2024) for F300M to scale and subtract stellar emission from F770W. For the longer-wavelength MIRI filters, we follow J. Sutter et al. (2024) and use spectral energy distribution models produced by the Code for Investigating Galaxy Emission (CIGALE; M. Boquien et al. 2019) to determine the factor to scale F300M to predict starlight emission in the filter. The formulae that we use are

We use the intensity in each filter in MJy sr⁻¹, the “ss” subscript indicates the starlight-subtracted intensity. We include the ±1σ uncertainty in starlight subtraction from the standard deviation from the CIGALE models computed over a variety of plausible CIGALE models (see J. Sutter et al. 2024) in each MIRI filter. Starlight subtraction makes the largest difference in regions with low MIRI intensities and high F300M, such as near-IR bright bulges and star formation deserts (see Section 2.3). Our sample includes 13 galaxies with such old stellar bulges and bars with very high surface densities of old stellar populations. In these regions, we find that stellar continuum accounts for ≲30% of the flux in F770W and as much as ∼5% of the flux in F2100W. In such regions, the starlight subtraction can represent the dominant source of uncertainty.

2.1.3. Estimating

With starlight-subtracted mid-IR maps, we measure the mid-IR color of F770W_ss to F2100W_ss across the full sample. F770W_ss/F2100W_ss is often employed to trace the PAH fraction in normal star-forming galaxies (e.g., J. Sutter et al. 2024; O. V. Egorov et al. 2025), following similar use of the 8/24 μm ratio with Spitzer (e.g., C. W. Engelbracht et al. 2005; F. Galliano et al. 2018; A. Li 2020). We estimate the mid-IR PAH-to-continuum emission ratio using the empirical calibration³² from J. Sutter et al. (2024) as

Note that while J. Sutter et al. (2024) and O. V. Egorov et al. (2025) did not subtract starlight from F2100W, we use F2100W_ss in . As noted above, this results in a discernible ≲5% effect only in extremely near-IR bright old bulges and star formation deserts.

Note that while we remove the stellar continuum, we do not subtract dust continuum emission from PAH-dominated filters F770W_ss and F1130W_ss since adjacent mid-IR filters are not available to reliably estimate the dust continuum in PAH filters. In previous work, the hot dust continuum is estimated to be ∼5×  lower than the 7.7 μm PAH feature contribution to F770W_ss (C. M. Whitcomb et al. 2023; D. Baron et al. 2024). Hence, we calculate other band ratios including F770W_ss/F1130W_ss (tracing PAH charge) without dust continuum subtraction. For consistency, we also compare our results to dust models in which we include the continuum emission. Future work will present improved dust continuum constraints for MIRI filters for PHANGS galaxies (L. Hands et al. 2026, in preparation).

In Figure 1, we show the diversity in resulting maps of F300M, F770W_ss, F2100W_ss, and for a few representative galaxies. All intensity maps share the same colorbar, and similarly all panels share the same colorbar for visual comparison between filters and across galaxies.

Zoom In Zoom Out Reset image size

Previous works have found mid-IR colors to vary with radiation field intensity and hardness (e.g., J. Sutter et al. 2024; D. Baron et al. 2025; O. V. Egorov et al. 2025). We use maps of SFR and specific star formation rate (sSFR≡SFR/M_*) as tracers of these factors. SFR and sSFR reflect the presence and relative abundance of young massive stars, which produce the majority of FUV photons in most star-forming galaxies and so drive changes in radiation field intensity and spectrum. We measure the sSFR using optical integral field unit (IFU) spectroscopy from the Multi Unit Spectroscopic Explorer (MUSE) on the Very Large Telescope (VLT) available for 42/71 of our galaxies. This includes the 19 Cycle 1 galaxies with MUSE data from the PHANGS-MUSE Large Program (1100.B-0651, PI: Schinnerer; E. Emsellem et al. 2022). The PHANGS-MUSE data coverage for the 19 galaxies extends out to ∼2 × R_eff or twice the effective radii, which is generally larger than the PHANGS-JWST footprint. MUSE data on an additional 23 galaxies are available from a combination of several small programs and archival data at ESO (see Table 1 in O. V. Egorov et al. 2025, for details of specific archival observations). This constitutes the “MUSE Extended” sample, and brings our total sample with existing MUSE data up to 42. Within the MUSE Extended sample, roughly half of the galaxies have coverage comparable to PHANGS-MUSE (full optical disk), while coverage for the other half is restricted to single central pointings (usually ∼1R_eff).

We use the “convolved and optimized” (copt) resolution MUSE data, which achieve a uniform Gaussian point-spread function (PSF) for each galaxy (as in E. Emsellem et al. 2022). The angular resolution of the MUSE cube varies by galaxy and is for our sample. We treat these data as effectively beam-matched to the JWST working resolution, since we only use them for SFR and sSFR measurements across azimuthal bins or morphological environments, i.e., regions much larger than a resolution element.

As in E. Emsellem et al. (2022), both the PHANGS-MUSE and MUSE Extended sample are reduced using the PHANGS-MUSE pipeline pymusepipe³³ to produce data cubes, and the PHANGS data analysis pipeline DAP³⁴ to produce maps of SFR and stellar mass (see E. Emsellem et al. 2022, for details). DAP uses the MUSE Balmer decrement extinction-corrected Hα maps to trace the star formation surface density, Σ_SFR (E. Emsellem et al. 2022; F. Belfiore et al. 2023a), and models the stellar continuum (e.g., I. Pessa et al. 2022, 2023) to trace the stellar mass surface density, Σ_*, which are reliable on ≳80 pc scales, and certainly on environment-integrated scales, but can break down on smaller scales in ≲50 pc-scale H ii regions or high-attenuation centers (E. Emsellem et al. 2022). The ratio of the two is sSFR = Σ_SFR/Σ_*. When measuring sSFR in large regions, we sum total SFR and stellar mass in the region before dividing.

For the 32 galaxies in our sample without MUSE coverage, we use F2100W and F2100W/F300M as proxies for Σ_SFR and sSFR. We expect this approach to be less accurate than the MUSE products, but it has the advantage of being available for all targets. We discuss this more in Section 2.4.

We measure the variation in mid-IR colors as a function of morphological environment. We use the M. Querejeta et al. (2021) environmental masks that define galaxy centers, bars, and disk regions based on Spitzer IRAC near-IR imaging (see also H. Salo et al. 2015). Within each environment (center/bar/disk), we calculate the integrated Σ_SFR and sSFR as described above. Based on this, we identify four different types of environments³⁵ :

1. 1.  
   Old stellar population-dominated star formation deserts or bulges. Galaxy centers or bar environments (inter-bar regions) with low sSFR (, or sSFR ≲ 7 × 10⁻¹² yr⁻¹), hereby referred to as “bulges.”
2. 2.  
   Young star-forming centers. Galaxy centers and CMZs³⁶ with high Σ_SFR (median F2100W intensity>10 MJy sr⁻¹, or Σ_SFR ≳ 0.03 M_⊙ yr⁻¹ kpc⁻²).
3. 3.  
   Normal centers. Galaxy centers with Σ_SFR and sSFR typical of the rest of the galaxy.
4. 4.  
   Disks. All environments not classified as a galaxy center or “bulge,” with Σ_SFR and sSFR typical of the rest of the galaxy. This includes, e.g., weak bars, spiral arms, and inter-arm regions.

While we use the M. Querejeta et al. (2021) morphological masks to distinguish between galaxy centers, bulges, and disks as our fiducial environment selection, we note three special cases. In NGC 4826, our MIRI and NIRCam observations cover the central ≈2ℓ_*, where ℓ_* is the exponential scale length of the stellar disk (H. Salo et al. 2015; A. K. Leroy et al. 2021).³⁷ NGC 4826 is close to an early-type galaxy (ETG; SAab Hubble type), where the inner ℓ_* is star-forming with high gas column density within an early-type disk galaxy where a “recent” merger likely caused the counter-rotating H I disk dominated by older stellar populations at R_gal ≳ ℓ_*. Hence, in NGC 4826, in addition to the M. Querejeta et al. (2021) environment masks, we classify the outer >ℓ_* of the galaxy as a star formation desert. Finally, NGC 3368 and NGC 3344 were not included in M. Querejeta et al. (2021). Lacking environment masks, we visually classify the central ℓ_* as the “galaxy center” (ℓ_* measurement from A. K. Leroy et al. 2021), and r > ℓ_* as the “disk” environment for each galaxy.

For the 32 galaxies without MUSE coverage, we use the median F2100W intensity in each environment as a Σ_SFR indicator. We also use the ratio of median F2100W-to-median F300M as a proxy for sSFR. MIRI F2100W is analogous to widely used IR SFR tracers including WISE4 and Spitzer MIPS 24 μm (e.g., R. C. Kennicutt & N. J. Evans 2012; T. H. Jarrett et al. 2013; A. K. Leroy et al. 2019) and has been explored as an SFR tracer using JWST including by F. Belfiore et al. (2023b); D. Calzetti et al. (2025). NIRCam F300M is analogous to the commonly used M_* tracers WISE1 or IRAC1 (e.g., S. E. Meidt et al. 2014; M. Querejeta et al. 2015; A. K. Leroy et al. 2019) or the widely used K band (e.g., T. H. Jarrett et al. 2003; M. F. Skrutskie et al. 2006).

In Figure 2, we show how F2100W and F2100W/F300M correlate strongly with Σ_SFR and sSFR for the 42 galaxies with MUSE coverage. The plot indicates strong correlations of Σ_SFR with F2100W (Spearman rank correlation coefficient ρ = 0.89) and sSFR with F2100W/F300M (ρ = 0.82) for the 42 galaxies with MUSE and JWST measurements, with < 0.5 dex scatter.³⁸ This supports our use of these IR intensities and colors as proxies for star formation activity where MUSE data are not available.

Zoom In Zoom Out Reset image size

Using these tracers, we select young star-forming centers as galaxy center regions within the M. Querejeta et al. (2021) mask where the median F2100W intensity is >10 MJy sr⁻¹. This translates to roughly Σ_SFR > 0.03 M_⊙ yr⁻¹ kpc⁻². We select old starlight-dominated bulges and star formation deserts as galaxy center and bar regions with , which translates roughly to sSFR≲7 × 10⁻¹² yr⁻¹. Both cuts are indicated in Figure 2 as dashed horizontal lines.

2.5.1. Constructing Radial Profiles

We measure the radial variation in dust emission, mid-IR colors, and in the diffuse ISM in each galaxy. To consistently compare galaxies of different sizes and morphologies, we construct radial profiles in azimuthal bins of width 0.1ℓ_*, the exponential disk scale length. For reference, in a simple exponential disk, we expect that the effective radius relates to the scale length as R_eff = 1.68ℓ_*. However, the radial profiles of many of our galaxies cannot be fit by a single exponential due to the presence of CMZs or bars. We thus prefer ℓ_* instead of the R_eff or a half-light radius measured in the near-IR, since many of our galaxies show strong nuclear emission from active galactic nucleus (AGN) disks or nuclear stellar disks. In these cases, R_eff is affected by this bright compact emission, which results in smaller R_eff values, and is not a good predictor of the large-scale disk structure.

To construct radial profiles, we calculate the median intensities in each azimuthal 0.1ℓ_* wide bin (Section 2.3). We use the median³⁹ to suppress the contribution of H ii regions and properly access the diffuse ISM values. H ii regions are bright, but occupy only a small fraction of the total area in any radial bin or environment mask (see Section 2.5.3 for exceptions), and typically represent only the tail end of the distribution of pixel intensities (D. Pathak et al. 2024). We demonstrate in Figure 3 that the ratio of medians is relatively insensitive to whether nebular regions are masked. We include the radial profiles of NGC 1300 since it is particularly illustrative—NGC 1300 includes all three major environments (young center/star formation desert/disk; Section 2.3), which are also discernible from the radial profile.

Zoom In Zoom Out Reset image size

We measure the statistical uncertainty around the median using the error extension for JWST images (T. G. Williams et al. 2024) appropriately adjusted for the number of independent resolution elements in the bin. This statistical uncertainty is typically <5%. The backgrounds are uncertain to ∼0.03–0.09 MJy sr⁻¹, depending on filter (see Section 2.5.4), and this uncertainty is covariant across the entire profile.

We clip the profiles where the median intensity falls below 5×  the propagated 1σ error in the median in that bin, or falls to within 10% of the background intensity in a filter (e.g., where F2100W falls below 1.1 × 0.09 ≈ 0.1 MJy sr⁻¹). This typically rejects radial bins in the outskirts where the emission is faint and the background is uncertain, such as corners of the mosaic. We also clip any bin at large R_gal that contains <50 independent resolution elements, which again rejects corners of images where not enough azimuthal extent is covered in a bin. Finally, for galaxies with inclination >45° we do not include emission within ±30^∘ of the minor axis when binning by radius, which is sufficiently conservative since the binned medians and percentiles are robust to <10% for ±10^∘ variation in the azimuthal cut along the minor axis. This avoids line-of-sight confusion between the vertical and radial distributions of dust.

2.5.2. Measuring Mid-IR Colors of Diffuse Emission

Within azimuthal bins and kiloparsec-scale environments, we measure mid-IR colors. The ratio of medians is more resistant to outliers than the median of pixel-wise ratios, particularly in the faint, diffuse parts of disks. Therefore, we report all colors (intensity ratios) as

In Figure 3 (middle), we show that the radial profiles of the ratios of medians are generally in agreement with the medians of pixel-wise ratios. We confirm this holds true for most radial bins at R_gal ≤ 3ℓ_*.

We also check whether our median-based ratios effectively pick out diffuse emission. Our primary concern is that our measurements might be affected by emission from small but bright H ii regions where PAHs can be destroyed. To test this, we compare radially averaged bins of environment-integrated colors with and without masking nebular regions. We use the MUSE nebular region masks where available (B. Groves et al. 2023; O. V. Egorov et al. 2025), and otherwise denote all pixels with F2100W intensities >3 MJy sr⁻¹ as nebular, since the locations of H ii regions typically show high F2100W intensities (D. Pathak et al. 2024).

In Figure 3 (right), we show that there is indeed good consistency between mid-IR colors with or without masking nebular regions. Since the nebular regions are only available for a subset of galaxies with MUSE data, and the MUSE-identified nebular masks are often larger than the actual physical size of the nebulae and include significant diffuse emission in addition to the central ∼10 pc-scale H ii regions (see, e.g., A. T. Barnes et al. 2026), we prefer to treat the median as our fiducial measurement. This allows us to provide measurements that include all pixels in an environment while still reflecting diffuse emission in the ISM.

2.5.3. F2100W in Galaxy Centers

Young star-forming centers often host CMZs or AGN, which can saturate or contaminate the emission with strong diffraction spikes in F2100W. We visually inspect each galaxy center and remove from analysis the saturation- or diffraction artifact-dominated centers in NGC 1566, NGC 1637, NGC 3507, NGC 3627, NGC 4457, NGC 4597, NGC 4569, NGC 4941, NGC 5643, NGC 6300, and NGC 7496. We include an example of a center removed from the analysis in Figure 4.

Zoom In Zoom Out Reset image size

Star-forming centers also host CMZs and nuclear starbursts, where, due to the high spatial density of nebular regions, the median no longer reliably excludes nebular regions. Even for galaxies with MUSE nebular catalogs, the MUSE data lack the resolution to distinguish H ii regions from the surrounding medium. Instead, in the star-forming centers we separate nebular and diffuse emission using the JWST data. Correlating pixel-wise F2100W_ss with F770W_ss and within each center, we fit a broken power law to binned medians (similarly to R. Chown et al. 2025). Each broken power-law fit yields x_b, the value of x, which corresponds to the break in the power law. Whenever F2100W_ss > x_b, we consider the pixel nebular, where low F770W_ss and reflect PAH destruction in ionized gas, and we consider all pixels <x_b as diffuse. In Figure 5, we include an example galaxy center with the resulting selection for diffuse versus nebular pixels in NGC 1512.

Zoom In Zoom Out Reset image size

In our analysis, we show two measurements for CMZs wherever possible—filled symbols for all pixels in centers (diffuse + nebular), and empty triangles for diffuse-only emission identified using this broken power-law approach. We take the diffuse measurement as our fiducial value for comparisons with other environments.

2.5.4. Uncertainties

In Figure 6, we compile the radial profiles of median F300M, F770W_ss, and F2100W_ss for all 71 galaxies, and F1000W_ss and F1130W_ss for 20 galaxies following Section 2.5.1. We first show the profiles of each galaxy colored by M_⋆ and also show median profiles after binning the galaxies by global M_*. More massive galaxies (M_* > 10¹⁰ M_⊙) show the imprint of central bulges and bars in their steeper or flattened central radial profiles (departures from an exponential disk) in addition to higher overall intensities, compared to lower-mass galaxies which show flatter profiles (E. Emsellem et al. 2026).

Zoom In Zoom Out Reset image size

In Figure 7, the profiles of show systematic variation with galaxy M_*. The amount and scale of suppression in centers varies by galaxy , where the most massive galaxies (M_* > 10¹⁰ M_⊙) on median show the lowest in centers. The central in these massive galaxies is on median 0.1–0.2 dex lower than the nearly flat median beyond ℓ_* (discussed further in Section 3.3).

Zoom In Zoom Out Reset image size

remains nearly constant across most radial bins in the star-forming disks of “normal” galaxies beyond ∼1ℓ_*. This is reflected in both the individual and stacked radial profiles of , which remain approximately flat on average at ℓ_* < R_gal < 3ℓ_* for most galaxy disks (discussed further in Section 3.2).

Our NIRCam footprints typically extend out to only ∼2.5–3ℓ_*, and rarely cover significant area beyond 4ℓ_*. Figure 7 shows that the profiles of and remain flat over this range, but our observations do not cover the outskirts of star-forming disks, where is expected to vary due to changes in metallicity, sSFR, and lower column density (see, e.g., G. Aniano et al. 2020; C. M. Whitcomb et al. 2024; R. Chown et al. 2025). We thus do not cover enough radial extent and range in galaxy M_* and metallicity to recover any systematic trend in due to changing metallicity in the diffuse ISM.

Table 1 summarizes our measurements of , which is defined in Equation (1). The radial profiles of for most galaxies show a stable plateau between roughly 1 and 3ℓ_*. In Figure 8, we show this plateau value as the median over all 0.1ℓ_* wide radial bins with ℓ_* ≤ R_gal ≤ 3ℓ_*, as well as the 1σ scatter across azimuthal bins. We check against global (galaxy-integrated) properties (from A. K. Leroy et al. 2021) for correlations between the plateau value of . We find that the plateau value correlates best with global , with a moderate Spearman rank correlation coefficient of ρ = 0.42 (p-value< 0.0001). While galaxy sSFR presents the strongest correlation, we also note that the lowest galaxies show the lowest (no overall correlation with : ρ = −0.11, and p-value =0.372).

Zoom In Zoom Out Reset image size

The diffuse median at ℓ_* ≲ R_gal ≲ 3ℓ_* is in agreement with the previously measured diffuse ISM median () (using F2100W instead of F2100W_ss) for 19/71 of our galaxies by J. Sutter et al. (2024), shown as the black dotted–dashed line in Figures 7 and 8. Although on median the profile of the lowest mass bin () is in line with the rest of the sample (see Figure 7), in the lowest- galaxies in our sample with , remains universally low in the entire disk relative to more massive galaxies (Figure 8). In particular, our lowest mass galaxy is NGC 3239 (included in Figure 1) with M_* ≈ 10^9.17 M_⊙, and one of the lowest . Table 1 summarizes the median and scatter in across all radial bins for 71 galaxies. Future work integrating our results with JWST observations of lower-mass dwarf galaxies will be useful to test how this trend transitions into the well-known suppression of PAH abundance in low-mass dwarf galaxies (e.g., F. Galliano et al. 2018; A. Li 2020).

In addition to radial profiles, we split each galaxy into distinct morphological and star formation environments (Section 2.3). Table 2 reports the median and spread in mid-IR colors for each environment,⁴⁰ and in Figure 9, we plot and two other key ratios for each environment in each galaxy. We compare these mid-IR colors to the average Σ_SFR, sSFR, and F2100W/F300M color integrated over each environment. We indicate “normal” disk environments with yellow markers and “normal” centers in pale blue (X’s if all four MIRI filters are available, circles otherwise).

Table 2. Summary of Diffuse Mid-IR Colors

Environment	Median	Q1/Q3	±1σ
Normal Disk (20)	−0.24	−0.28/ − 0.21	0.08
Normal Center (8)	−0.26	−0.28/ − 0.25	0.05
Young CMZ (9)	−0.18	−0.21/ − 0.13	0.05
Old Bulge (4)	−0.43	−0.47/ − 0.40	0.04
Normal Disk (20)	−0.24	−0.26/ − 0.20	0.06
Normal Center (8)	−0.19	−0.21/ − 0.17	0.06
Young CMZ (9)	−0.40	−0.50/ − 0.35	0.10
Old Bulge (4)	−0.17	−0.21/ − 0.14	0.04
Normal Disk (20)	0.11	0.05/0.16	0.09
Normal Center (8)	0.10	0.05/0.11	0.05
Young CMZ (9)	−0.03	−0.07/0.03	0.07
Old Bulge (4)	−0.10	−0.13/ − 0.07	0.04
Normal Disk (20)	0.35	0.32/0.38	0.05
Normal Center (8)	0.36	0.27/0.38	0.08
Young CMZ (9)	0.08	0.01/0.20	0.11
Old Bulge (4)	0.34	0.31/0.37	0.05

Note. Median, first/third quartiles, and standard deviation of mid-IR colors in the diffuse ISM of the n = 20 galaxies with full MIRI filter coverage (Section 2.1), split by galaxy environment (Section 2.3).

Download table as:  ASCIITypeset image

Zoom In Zoom Out Reset image size

Across our full sample, most disks show a narrow range of values in (similar to Figure 8), with <0.1 dex scatter (Table 1). Figure 9 also shows a subset of galaxy centers that have similar environment-integrated Σ_SFR and sSFR to normal disks and also show similar mid-IR colors. “Normal” disks and “normal” centers also show a narrow range of values around and , with environment-integrated medians and scatter summarized in Table 2.

As with the radial profiles (Figure 8), Figure 9 shows a moderate correlation between environment-integrated sSFR and in “normal” disks (ρ = 0.47). We also find a strong correlation between F770W_ss/F1130W_ss and sSFR (ρ = 0.86), or F2100W/F300M (ρ = 0.84), which we use as a proxy for sSFR (Section 2.4), with best-fit relations reported in Figure 9.

Figure 7 shows that declines in the centers of massive galaxies. The most massive galaxies (M_* ≳ 10¹⁰ M_⊙) show kiloparsec-scale low . Particularly within the central ∼1ℓ_* (typically ∼3 kpc), the depression in is significant, 0.1–0.4 dex lower relative to the rest of the disk (Figure 7). Examples include the central regions of NGC 1300 (Figure 3), NGC 1512, and NGC 2775 (Figure 1).

As discussed in Section 2.5.4, we expect uncertainties associated with starlight subtraction to dominate in galaxy centers with faint PAH emission. Figure 9 shows our best-estimate error bars accounting for these uncertainties. The figure shows that the contrasts between (i) bulges and disks and (ii) young centers and disks are both larger than the uncertainty associated with starlight subtraction.

Local conditions in inner galaxies can diverge from those in normal star-forming disks in different ways. CMZs are expected to have increased radiation field intensity because of the high density of young, massive stars. This results in a higher than normal density of H ii regions where ionizing photons may destroy fragile PAHs (e.g., J. Sutter et al. 2024; O. V. Egorov et al. 2025, for in H ii regions). It also enhances the 21 μm continuum (e.g., G. J. Bendo et al. 2008). CMZs can also exhibit higher average dust attenuation (1 ≲ A_V ≲ 3 mag; J. D. T. Smith et al. 2007, see Section 4.5), total gas column (M. Querejeta et al. 2021), and stellar mass density than disks.

The centers of massive galaxies can also host quiescent old stellar population-dominated bulges and bars, where the radiation field is softer, i.e., FUV-poor, due to the low density of young stars and high density of older stars (e.g., the FUV-poor bulge of M31; B. Groves et al. 2012; B. T. Draine et al. 2021). These conditions may be more similar to ETGs and ellipticals (e.g., H. Kaneda et al. 2005; R. Rampazzo et al. 2013). Bulges can also be relatively ISM-free regions where the total gas column density and dust content can be lower than in normal star-forming disks.

Finally, though we have masked the brightest cases (Section 2.5.3), 5 of our galaxies also host active galactic nuclei (AGN; M.-P. Véron-Cetty & P. Véron 2010) which can change PAH and mid-IR continuum emission properties (J. D. T. Smith et al. 2007; J. J. Jensen et al. 2017) over the central ∼500 pc, the exact extent of the zone of influence around AGN remains poorly constrained.

These changes in radiation field intensity and shape can change by changing the intensity of the dust continuum, the ionization state or size distribution of PAHs, and/or the importance of stochastic emission versus equilibrium emission from small grains (D. Baron et al. 2025). These, in turn, manifest in the observed mid-IR colors. In Figure 9 and hereafter, we measure these colors for each environment in each target galaxy. We show variations as a function of environment-integrated Σ_SFR and sSFR. Σ_SFR correlates with increasing radiation field intensity (e.g., J. Chastenet et al. 2025). It is highest in starburst galaxy centers and lower in normal star-forming disks and bulges. On the other hand, sSFR normalizes the overall star formation surface density by the existing M_*, which should correlate with the shape of the radiation field. sSFR is low in FUV-poor bulges and star formation deserts, but higher in normal star-forming disks and galaxy centers.

Figure 9 shows that there is significant variation in diffuse mid-IR colors among environments, and that these mid-IR color variations correlate with Σ_SFR and sSFR. While remains roughly constant in the diffuse ISM of most normal disks (intermediate Σ_SFR and sSFR), it decreases in extreme environments—(1) in young CMZs with very high Σ_SFR, and (2) bulges with very low sSFR. Such large variations in local star formation result in changes in both the intensity and shape of the radiation field.

3.3.1. Young Star-forming Centers

3.3.2. Old Bulges and Star Formation Deserts

For the 20 galaxies where all four mid-IR filters are available, we show mid-IR colors as a function of one another to connect the trends seen in Figure 9. In Figure 10, we show F770W_ss/F2100W_ss (or ) and F770W_ss/F1130W_ss PAH-correlated ratios against the mid-IR continuum ratio F1000W_ss/F2100W_ss. Comparing F770W_ss/F2100W_ss and F1000W_ss/F2100W_ss, it is clear that in most CMZs and disks, 7.7/21 μm and 10/21 μm vary together. If F1000W_ss is taken to reflect the continuum near F770W_ss, this indicates that to first order, mid-IR color variations contrasting “normal” disks versus CMZs are likely primarily driven by variations in the radiation field intensity, not changes in PAH properties. We show in Section 4.3 that environments with low 7.7/21 μm and 10/21 μm also show higher far-infrared (far-IR) dust temperature (and hence, radiation intensity), supporting this interpretation.

Zoom In Zoom Out Reset image size

Figure 10 also highlights the bulges as distinct environments that show abnormally low 7.7/21 μm despite normal-to-high 10/21 μm. This departure from the otherwise consistent trend between centers and disks indicates that the low F770W_ss in bulges is likely not due to changes in only radiation field intensity. The right panel of Figure 10 compares F770W_ss/F1130W_ss and F1000W_ss/F2100W_ss. This once again highlights the contrast between highly star-forming CMZs with high F770W_ss/F1130W_ss, and quiescent bulges with extremely low F770W_ss/F1130W_ss, lower than normal disks. In Section 4.4, we confirm that changes in PAH charge and radiation hardness likely suppress 7.7 μm emission and result in low in quiescent bulges.

The mid-IR colors of the diffuse ISM in star-forming galaxies vary between “normal” galaxy disks, highly star-forming galaxy centers, and quiescent bulges or star formation deserts. Mid-IR colors in the diffuse ISM on average remain relatively constant across large parts of the optical disks of galaxies, with <0.1 dex variation in colors across large samples of galaxies (Table 2).

Highly star-forming CMZs and quiescent bulges both harbor conditions very different from normal star-forming disks. In CMZs, the average gas and dust column density, and hence A_V, can be high. In addition, the high density of young, massive stars leads to a high radiation field intensity and can also result in more ionizing photons that can destroy PAHs and result in harder (FUV-rich) radiation fields (e.g., L. E. Tacconi-Garman et al. 2005; O. V. Egorov et al. 2025), which can enhance dust continuum emission, suppress PAH emission where PAHs are destroyed, or may enhance emission from more ionized, and/or smaller, and/or shorter-wavelength PAH emission. In Section 4.3, we confirm that the variation in mid-IR colors in CMZs is likely driven by changes in the overall radiation field intensity in the diffuse ISM.

On the other hand, quiescent bulges, bars, and star formation deserts are characterized by high stellar surface densities, but can be relatively ISM-free. Although the radiation field intensity can be moderately high in bulges, the radiation spectrum is expected to be FUV-poor (i.e., softer) due to older starlight (e.g., the high radiation intensity but soft, FUV-poor bulge of M31; B. Groves et al. 2012). Softer radiation can suppress short-wavelength PAH emission relative to longer-wavelength features (e.g., B. T. Draine et al. 2021), and may suppress emission from more ionized PAHs. In Section 4.4, we discuss how mid-IR color variations in relatively quiescent bulges may be due to the presence of more neutral PAHs and/or a softer radiation spectrum.

In addition to the actual abundance of PAHs, q_PAH, the physical properties of PAHs or the properties of the illuminating radiation field (or a combination of the two) may drive changes in mid-IR colors. In the rest of the paper, we try to constrain how much of the variation in mid-IR colors between environments can be explained by local conditions such as changes in the radiation field (intensity and spectrum) and/or PAH properties (charge and size), rather than q_PAH. B. T. Draine et al. (2021, hereafter D21) model the effect of each of these factors, and we compare D21 dust model predictions and our mid-IR photometric filter ratios. We use a fixed Milky Way q_PAH throughout, reflecting the relatively high metallicity of our targets, to highlight the range of variation possible due to radiation field and PAH properties only. We compare against three radiation field hardness models from D21 which roughly correspond to the three environments we classify (Section 2.3)—a 3 Myr old (young) starburst (blue; labeled as SB3), a modified Mathis, Mezger, and Panagia (mMMP) radiation field (J. S. Mathis et al. 1983) that roughly corresponds to a Milky Way disk average (yellow; mMMP), and the quiescent Andromeda bulge (red; M31 bulge). These radiation models have decreasing radiation hardness, from most FUV photon-rich (SB3) to most FUV photon-poor (M31 bulge). We note that we use the D21 models directly and produce grids (variation due to changing PAH charge and size) for a given U, instead of a range of U. While a range of U is likely a more realistic assumption for dust on larger scales, lacking finer constraints, we compare our observations against a range of different (single) U values. We show D21 predictions for the fiducial diffuse ISM models for comparison, and include for reference the predictions for Astrodust+100% neutral PAHs where relevant.

We perform synthetic photometry (see, e.g., J. Koornneef et al. 1986; K. D. Gordon et al. 2022) using the transmission curves for the respective JWST filters on the D21 model spectral energy distributions (SEDs) for varying radiation field intensity U, three radiation hardness models (SB3, mMMP, and M31 bulge), and varying PAH ionization state and grain size distributions. We use the model SEDs that include both PAH emission and dust continuum. This enables consistent comparison with our measurements, where we remove starlight but not dust continuum emission from MIRI filters.

In Figure 11, we include D21 model grids that show the range of color combinations predicted by varying PAH ionization and size for the three radiation hardness models at fixed , and include the SB3 grid at to highlight the direction of change with . As noted in Section 3.4, the lower F770W_ss/F2100W_ss and lower F1000W_ss/F2100W_ss in CMZs compared to disks is in good agreement with variation predicted due to increasing in D21.

Zoom In Zoom Out Reset image size

Figure 11 shows the observations to be offset from the model grid. This is because the F1000W_ss intensities that we measure are systematically higher than those predicted by the D21 models by 0.1 dex. We identify the F1000W_ss filter as the cause of the 0.1 dex discrepancy, because our observed F770W_ss/F2100W_ss colors overlap the range predicted by the models, as do the F770W_ss/F1130W_ss and F1130W_ss/F2100W_ss colors (shown below, see, e.g., Section 4.4). The bulges also show significantly lower F770W_ss/F2100W_ss relative to all other environments, which are not reproduced by the standard range of PAH charge variation in the D21 models, but are still bracketed by the limiting case of Astrodust+100% neutral PAHs (darker lines labeled PAH⁰ in Figure 11), as discussed in Section 4.4.

The offset between D21 and our measurements of F1000W persists across all environments in our sample. We speculate that the discrepancy in F1000W may be due to a combination of (a) the adjacent PAH emission features in the models being too narrow, and so not contributing sufficiently to the F1000W filter; or (b) the shape of the continuum emission around 10 μm being different in the models. The D21 SEDs do not include differential attenuation of the IR emission, e.g., by the 10 μm silicate absorption feature, so the low F1000W in the models cannot be due to missing silicate absorption. In fact, if we were to account for the silicate absorption in the models, this would further decrease the F1000W flux from the models and hence would make the discrepancy between the models and the observations even larger (detailed SED comparison forthcoming in M. Boquien et al. 2026, in preparation).

At high radiation intensity (), a component from dust in thermal equilibrium starts to contribute at the longer MIRI wavelengths, leading to a relative increase in 21 μm emission. Because includes F2100W_ss, such intense radiation fields can cause variations in mid-IR color that resemble those due to varying PAH abundance. To check whether the low in highly star-forming CMZs is due to high U or changes in q_PAH, we require an independent estimate of U. We use far-IR colors, which capture variations in the dust temperature, T_dust. In turn, T_dust reflects thermal equilibrium between the dust and the ISRF, and so U.

For 31 galaxies, we use archival Herschel PACS 70, 100, and 160 μm data compiled by J. Chastenet et al. (2025) to measure 70/100 μm, 100/160 μm, and 70/160 μm colors. Though available at much lower resolution () compared to the JWST observations, these far-IR colors are robust tracers of thermal dust temperature (T_dust), which is not accessible with JWST alone. For a single radiation field of intensity U and a modified blackbody dust SED with emissivity index β = 2, T_dust ∝ U^1/6. However, similar to the mid-IR (see Section 4.2), far-IR colors can depend on the distribution of radiation field intensities, so the interpretation in terms of a single T_dust or U is not straightforward, and we simply present the far-IR colors as a tracer of changes in the average radiation field.

In Figure 12, we show that high sSFR centers indeed show high 70/100 μm, 100/160 μm, and 70/160 μm colors, indicating high average T_dust, and hence significantly higher U relative to “normal” disks. In addition, some bulges also exhibit moderately high far-IR colors, and hence T_dust, despite their extremely low sSFR. This is likely due to their high concentration of old stars (high F300M intensities), similar to the bulge of M31 with high T_dust due to intense but old starlight (B. Groves et al. 2012).

Zoom In Zoom Out Reset image size

We complement this analysis using less direct but higher resolution mid-IR-based tracers of U. We expect that Σ_SFR and I_MIR/Σ_gas, particularly F2100W_ss/Σ_gas, should correlate with U. This follows since, to first order, in the diffuse ISM with sufficiently low U (U ≲ 10),

The smallest molecules, including PAHs, are stochastically heated and in this single-photon limit their emission is linearly proportional to U. At very low U ∼ 1, F2100W similarly traces stochastically heated grains, and is expected to scale linearly with U. At intermediate U grains are still stochastically heated, but multi-photon effects can cause the F2100W emission to become nonlinear as a function of U, and finally at larger U ≳ 10, grains have steady-state temperatures that scale roughly as U^1/6. Even where F2100W emission and the radiation field intensity are not expected to scale linearly, the relation between F2100W and U is monotonic for realistic ISM conditions (B. T. Draine & A. Li 2001, 2007). So for a fixed dust-to-gas ratio D/G, dust continuum-to-gas column, or F2100W_ss/Σ_gas ≈ F2100W_ss/Σ_Mol should scale with the radiation field intensity U. While there are caveats, like the need for an α_CO prescription to measure Σ_Mol, the possibility of D/G changes, and nonlinearity at high enough U ≳ 10 (where F2100W increases nonlinearly but still monotonically with U), we expect F2100W_ss/Σ_Mol to be a good tracer of relative changes in U.

In Figure 12, we test how F2100W_ss/Σ_Mol compares to far-IR colors by combining our F2100W_ss JWST measurements with molecular gas maps from PHANGS- Atacama Large Millimeter/submillimeter Array (ALMA; A. K. Leroy et al. 2021). We use the native ∼1″ resolution broad masked CO(2–1) maps, which allow for more complete flux retrieval in diffuse regions. We convolve F2100W_ss to match ALMA resolution, downsample both maps to the resolution of the ALMA beam, and then measure the median F2100W_ss/I_CO(2−1) ratio in each environment. We note that we fail to reliably detect some gas-poor bulges and faint disks in I_CO(2−1), and hence in Σ_Mol, despite having full-galaxy ALMA coverage, and exclude these low-confidence Σ_Mol regions from Figures 12 and 13. In addition, while the low resolution of Herschel means we cannot distinguish diffuse emission from overall emission for measuring colors, using the higher resolution F2100W_ss/I_CO(2−1) allows us to separate diffuse emission (as in Section 2.5.3).

Zoom In Zoom Out Reset image size

The correlation between F2100W_ss/I_CO(2−1) and far-IR colors is weak (ρ ≈ 0.25–0.4). However, much of the scatter appears to be driven by variations in the CO-to-H₂ conversion factor, α_CO, especially the contrast between dense, star-forming centers and more diffuse disks. In the second row of Figure 12, we use maps from J. Sun et al. (2025) to translate I_CO(2−1) into molecular gas surface density, Σ_mol. These maps are available for 57 of our galaxies, and use the E. Schinnerer & A. K. Leroy (2024) prescription for estimating , accounting for variations in excitation, emissivity, and metallicity.

The correlation between all three far-IR colors and F2100W_ss/Σ_Mol appears much stronger than with F2100W_ss/I_CO(2−1) (ρ ≈ 73). The strong correlation supports the use of F2100W_ss/Σ_Mol as a U indicator. The improvement with better α_CO treatment is also an indirect validation of the α_CO maps we use. Finally, we check that the use of F2100W_ss/Σ_Mol as an indicator of U marks an improvement over simply using MUSE Σ_SFR (which correlates strongly with F2100W, as shown in Figure 2).

We thus interpret the relative changes in F2100W_ss/Σ_Mol as relative changes in U. Then, from Figure 13, in the diffuse ISM of CMZs, U appears higher than that in most disks. Despite low sSFR, many old stellar bulges also have high U. This likely reflects the high intensity of old starlight, e.g., as seen in the bulge of M31 (B. Groves et al. 2012). Finally, most normal star-forming disks show low U.

At fixed PAH abundance and high U, the apparent depends on U because U affects the dust continuum emission at 7.7 and 21 μm differently. In Figure 13, we show how our measured , F770W_ss/F1130W_ss, and F1000W_ss/F2100W_ss vary with F2100W_ss/Σ_Mol, our rough indicator of U, as well as the expected variation in the same colors with varying U for three D21 radiation spectrum models. Comparing the observed trends with F2100W_ss/Σ_Mol and the model-predicted trends with U, we confirm that the low measured in CMZs is consistent with high local U, where F2100W_ss rises as a function of U, rather than an actual change in PAH mass fraction. The bulges show abnormally low for moderate to low (similar to disks) F2100W_ss/Σ_Mol, suggesting that the low in bulges is not due to high U. This interpretation is further supported by the similarity between the trend in F1000W_ss/F2100W_ss we measure across environments and the model curves for varying U. F1000W_ss/F2100W_ss shows a similar trend to , with a clear contrast between disks/bulges and CMZs—CMZs show 0.2–0.4 dex lower F1000W_ss/F2100W_ss than all other environments. Hence, the variation in F1000W_ss/F2100W_ss is likely also due to increasing U in CMZs.

Although changes in and F1000W_ss/F2100W_ss in CMZs are thus likely, the correlation between F770W_ss/F1130W_ss and F2100W_ss/Σ_Mol is likely not due to changes in U. As shown in Figure 13, the D21 models predict that F770W_ss/F1130W_ss is expected to remain relatively unchanged with U, and primarily encodes variation in PAH ionization, size, and radiation hardness. Indeed, most of the variation in F770W_ss/F1130W_ss at fixed U in the models reflects changes in ionization. This indicates that diffuse regions in the CMZs may host more ionized, and/or smaller PAHs, and/or harder radiation fields (high F770W_ss/F1130W_ss), and bulges likely host larger, and/or less ionized PAHs, and/or softer radiation fields (low F770W_ss/F1130W_ss). However, changing radiation field hardness and/or PAH size alone cannot explain the extent of the F770W_ss/F1130W_ss change we measure across environments (spanning roughly 0.5 dex). We further investigate the degeneracies between PAH properties (ionization, size) and radiation hardness in Section 4.4.

In quiescent bulges and star formation deserts, changes in U are not enough to explain observed colors. The bulges show unexpectedly low F770W_ss despite low-to-moderate U, and normal F1000W_ss/F2100W_ss. The low F770W_ss despite typical F1000W_ss and F2100W_ss in bulges can be due to three physical reasons:

1. 1.  
   Shape of the radiation field spectrum. PAHs may be illuminated by softer radiation fields in bulges due to the high density of old stars but little to no ongoing star formation. Soft (FUV photon-poor) radiation fields are expected to suppress short-λ PAH emission (e.g., D21).
2. 2.  
   PAH charge. F770W_ss primarily traces emission from the 7.7 μm PAH feature from more ionized PAHs. The presence of more neutral PAHs thus might explain reduced F770W_ss/F1130W_ss ratios in bulges.
3. 3.  
   PAH size. The presence of larger grains⁴¹ can suppress the 7.7 μm PAH feature relative to 11.3 μm.

To break this degeneracy between softer radiation, lower PAH ionization, and larger PAH size, we include NIRCam F335M emission available for the 19 Cycle 1 galaxies. We use maps of F335M_PAH from Koziol et al. (submitted), which isolate PAH emission from the F335M filter (including aliphatic emission) using a linear combination of the flanking F300M and F360M NIRCam filters to characterize and subtract the relevant stellar continuum, after correcting for any PAH contamination in F360M. Both IR spectroscopy and imaging of metal-rich ISM regions (comparable to our sample) typically exhibit weaker 3.3 μm PAH emission compared to the longer-wavelength PAH features at 7.7 and 11.3 μm (e.g., D. A. Dale et al. 2023; A. Maragkoudakis et al. 2026) We thus restrict any comparison that includes F335M_PAH (Figure 14, 15) to the subset of pixels within each environment where the fainter F335M_PAH is detected at matched angular resolution. In lower intensity environments, particularly in bulges and disks, we check that F335M_PAH is detected in at least 25% of the pixels in each environment. Hence, Figures 14 and 15 only include the amount of flux in matched pixels where we reliably detect F335M_PAH in addition to the previous MIRI filters. We note that realistic variations in A_V across environments can attenuate at most ∼10% and on median <4% of F335M emission, and do not impact our analysis (see Section 4.5).

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

With the inclusion of F335M_PAH we now have access to three PAH-tracing filters—F335M_PAH (from small, neutral PAHs), F770W_ss (from smaller, ionized PAHs), and F1130W_ss (from larger, neutral PAHs). In Figure 14, we present the PAH-to-continuum ratios split by environment for the 19 Cycle 1 galaxies for all three PAH filters, with the predicted filter ratios from D21 for three radiation hardness models at included for comparison (dashed horizontal lines). Once again, all three PAH-to-continuum ratios are low in the CMZs, consistent with higher U and enhanced F2100W_ss. However, in the bulges, only F770W_ss/F2100W_ss is low, while both F335M_PAH/F2100W_ss and F1130W_ss/F2100W_ss remain high, matching the typical values for normal disks. This rules out softer radiation fields and larger sizes, both of which should also suppress F335M_PAH/F2100W_ss. This indicates that the low F770W_ss in bulges and star formation deserts is primarily due to PAHs being less ionized. Figure 14 also shows how F770W_ss/F2100W_ss is the PAH-to-continuum ratio that correlates most strongly with local sSFR outside of CMZs, and F1130W_ss/F2100W_ss remains roughly constant across bulges and disks.

Finally, in Figure 15 we compare all environments for the 19 galaxies in PAH color–color space, with the corresponding D21 dust grids (see Section 4.2) at for comparison. Contrasting F770W_ss/F1130W_ss, F335M_PAH/F1130W_ss, and F335M_PAH/F770W_ss shows that most “normal” galaxy disks and centers scatter within the M31 bulge part of the D21 grids, while CMZs generally show higher F770W_ss/F1130W_ss, but still remain within the M31 bulge corner of the grids. Since the D21 models do not include differential attenuation, we speculate that if there is silicate extinction in CMZs, it may move CMZs relative to disks in this direction by primarily attenuating 11.3 μm. CMZs also have lower F335M_PAH/F770W_ss and F335M_PAH/F1130W_ss than most “normal” star-forming disks, which may indicate the presence of more ionized, larger PAHs in galaxy centers.

Figure 15 once again highlights how quiescent bulges and star formation deserts are unique environments that show anomalously low F770W_ss, and lie completely outside of the standard parameter space for PAH charge and size variation in the D21 models. Combined with F335M_PAH/F1130W_ss values (primarily tracing size variation, both from neutral PAHs) in bulges being comparable to many normal disks, this again reinforces the interpretation that the PAH population in bulges is likely more neutral. We note that the colors we measure in bulges are bracketed by the limiting case of the D21 models with 100% neutral PAHs (Astrodust + PAH⁰; maroon/gray/navy lines in Figure 15). Although a 100% neutral PAH population in star formation deserts is physically unlikely, the better agreement between observed colors in bulges and D21 predictions for neutral PAHs is again consistent with the interpretation that PAHs are likely more neutral in bulges.

Quiescent bulges in our otherwise “normal” star-forming galaxies may present similar ISM conditions as quiescent ETGs and elliptical galaxies, where lower (>500 pc) resolution spectroscopy surveys suggest changes in PAH properties in such extremely FUV-poor environments (e.g., R. Rampazzo et al. 2013). Although softer radiation fields can suppress short-λ PAH features including 7.7 μm, in the M31 bulge, the photometric 7.7 μm PAH emission is weaker but still detected (B. Groves et al. 2012; B. T. Draine et al. 2014, D21). However, the current range of PAH charge and side distributions included in the D21 model for any radiation spectrum or intensity fails to reproduce the extremely low F770W_ss we observe in quiescent bulges in PHANGS galaxies, where much larger variations in PAH ionization may explain the suppressed F770W_ss but normal F1130W_ss. We note that in many ETGs, while the 7.7 μm PAH feature is virtually undetected, the 11.3 μm PAH feature is detected (e.g., H. Kaneda et al. 2005; P. Panuzzo et al. 2011; R. Rampazzo et al. 2013, 2014). We speculate that similarly, in extremely low sSFR bulges, the 7.7 μm PAH feature may be severely suppressed, arguably due to significantly lower PAH charges than what the D21 grids currently allow for, in order to produce the extremely low F770W_ss we observe in low sSFR regions.

Moderately high A_V in CMZs. In addition to high U, CMZs and starburst centers are high-density environments where the total gas and dust column density is expected to be higher than in normal disks. In Figure 16, we show that the highest sSFR galaxy centers also host high optical attenuation A_V. We measure the -scale A_V using the Hα/Hβ Balmer decrement and the J. A. Cardelli et al. (1989) extinction curve to estimate E(B − V), and then assume a Milky Way average reddening R_V  =  3.1. In Figure 16, we present median A_V in environments. Since the medians are measured only at regions where Hβ is reliably detected, allowing for the Balmer decrement measurement, the A_V values in Figure 16 mostly trace the extinction toward the locations of H ii regions and their surrounding diffuse ionized gas.

Zoom In Zoom Out Reset image size

Most highly star-forming galaxy centers show higher attenuation (A_V ≈ 1.5–2.5 mag) compared to disks and quiescent bulges (A_V ≈ 0.5–1 mag). In high A_V environments, higher silicate absorption at 10 μm may suppress F1000W_ss intensities. In addition, due to the higher U (Section 4.3), CMZs also show elevated F2100W_ss. So, both high U and higher A_V can contribute to low F1000W_ss/F2100W_ss in centers (see, e.g., Figure 9). Note that the optical depth τ_λ = A_λ/1.086, and the observed intensity . As a result, at A_V = 1, roughly 7% and 3.5% of the F1000W and F2100W emission, respectively, is attenuated, and at A_V = 2, roughly 14% and 7% of the emission is attenuated (using extinction curves from K. D. Gordon et al. 2023).⁴² Even at the shortest wavelength relevant to our analysis (3.36 μm), roughly 4.5% and 9% of the F335M filter emission is attenuated at A_V = 1 and 2, respectively.

The expected decrease due to attenuation is small (≲10%) compared to the 0.4 dex decrease in F1000W_ss/F2100W_ss we observe in CMZs (e.g., Figure 13). Hence, although A_V is higher in highly star-forming centers, A_V is not high enough in PHANGS galaxy centers for silicate absorption to significantly attenuate F1000W_ss. This is analogous to results from Spitzer spectroscopy for SINGS galaxies, where A_V is rarely high enough (A_V ≳ 5 mag) for 9.7 μm silicate absorption to matter (J. D. T. Smith et al. 2007). The consistently low F1000W_ss/F2100W_ss values in highly star-forming centers of otherwise normal PHANGS galaxies are thus likely due to increasing F2100W_ss intensities due to higher average U (Section 4.3). The alternate y-axis translates V-band attenuation to the corresponding optical depth at 10 μm due to silicate absorption, assuming the empirical K. D. Gordon et al. (2023) extinction curve at a Milky Way R_V = 3.1. For all measured A_V even in CMZs (A_V ≈ 1.5–2.5 mag), the relevant optical depth at 10 μm is ∼0.1. So in the CMZs with the highest A_V, at most ∼15% of the F1000W emission is likely attenuated, which is a negligible effect for our sample.

AGN. The presence of strong AGN can also change mid-IR colors due to the changing radiation field and high attenuation (e.g., J. D. T. Smith et al. 2007; L. Zhang & L. C. Ho 2023; G. P. Donnelly et al. 2024). This can include strong continuum emission due to stronger radiation fields, weak PAH emission due to the photodestruction of PAHs, or strong 11.2 and 17 μm indicating larger PAH sizes closer to the locations of AGN. While several galaxy centers with AGN are removed from our sample due to saturation and strong PSF artifacts in F2100W (see Section 2.5.3), five galaxy centers with AGN are included in our analysis—NGC 1097, NGC 1365, NGC 1672, NGC 4303, and NGC 4826 (M.-P. Véron-Cetty & P. Véron 2010). These AGN show little to no visible signs of mid-IR color variations due to weak AGN, while most strong AGN show strong diffraction artifacts in F2100W, are saturated in galaxy centers, and hence removed from our analysis (see Section 2.5.3). The remaining five AGN galaxy centers show ≲0.1 dex lower F770W_ss/F2100W_ss (Figure 10), somewhat higher F770W_ss/F1130W_ss (Figure 15), and comparable F335M_PAH/F1130W_ss. So in the handful of AGN in our final sample of galaxy centers, we do not recover a statistically significantly change in mid-IR colors due to the presence of AGN.